{"schemaVersion":1,"releaseVersion":"0.21.115","publication":{"effortReferenceReviewHash":"82c3016914968f27b8c09cd41e801fc1230c7be43fab83ae7ec67b4dbc53d3ad","effortReferencePolicyHash":"ba8637819f078509de36f694ba295f6d87f3b11d1efae1e13ae2d7288e7fedc5","planMeteringHash":"2fb246ffe78b7d9c9e98c3d79d41491350d0a6a7f2e112222ff30e9b48f2d3f5","planValueFormulaVersion":"plan-value-v0.2","aaSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","throughputHistoryHash":"35db83ab861b25d07d6bb810551dade706e3379ad27656334e630c67cd210e0f","reviewedSpeedClaimsHash":"d57c526bc419daae9b472065c139bb89ca4f62609627f2099ce9f9619e6f52a6","portfolioRevision":"3a6e6c1e879d5e4139f6038164db16181b83b41dac9a88df847c1addda6912f3","publicGenerationEvidenceHash":"5eb8d1752853407a7ccec70487e64e0ae50b21edc4c937cc3faf2dd66e7a58d7","modelsdevSnapshotId":"modelsdev-604d89749477ce01cca3","codingRosterRevision":"2026-10-09.1","codingRosterContentHash":"be95532ac265d0ce9db29220672542b39de9d1d38f3c7d34387a23d0fedb0157","releaseIdentityRevision":"2026-10-12.2","releaseIdentityContentHash":"acf095f61e4307c8a69cb341eb5f046941259ee4fa01a1b3cc684ed14489d69c","defaultLineagePolicyVersion":"2026-10-09.1","defaultLineageContentHash":"650e430e30a8cd7abe99aa577de7f7924e7b5bb4194d9889988d2e6d7da60d09","planIndexAliasRevision":"2026-10-15.1","planIndexAliasContentHash":"0862a0474924422d709fcd8d6c65a65936ea8933c896845754fc29d2c2389443","planAccessRevision":"2026-10-15.1","planAccessContentHash":"b4d7f7b4ccefe09e3261fe7b484000c7188a4c81423fd206402b62d40fa8cfb9","publicPriceConfigHash":"c73bb3a812c9301508a1963d54ee97b58aa46798277fef9b910c292a24355488","proOfferingHash":"d19034a0de0a77933255f6a0b91b018af845113b884b62475e5719afaeb84720","reviewedPlanRegistryRevision":"2026-10-15.1","reviewedPlanRegistryContentHash":"4c8e28fb103fb84d15eabbd135f08b23e0618134f92effae72f699a97a8ed90c","planProviderRegistryContentHash":"bd8c9bb5573d9f163eded9da348f388dcca308a438fd7565f60ad8457056a471","planAliasContentHash":"451a4223145764e23d37489140220d82e4ccb2c23b164a2a9864462c08cdd71d","corePlanEvidenceHash":"cbd37dd93c1833ad2bf427b3e333ce39e0fb83bac1e8d58dbfb17e6e081d1026","planTierRegistryRevision":"2026-10-15.1","planTierRegistryContentHash":"514cc0c546c496e89410f2c718e4641ea3148bee0e284df8c0cfe1333fdd299c","toolUsageHistoryContentHash":"0948fb934d403c0cca5ccb7a249ad6325a8cdb9fd1d01b79eab85d809c56fa3c","corePlanRegistryContentHash":"b0c3c9260ec479da43b4d0a214dabf0f2791e87556120c991f90c1a9fcc67d47","planProductRevision":"2026-10-15.1","planProductContentHash":"9808ec0cab229ba780628783b438f293e08fe08a7fef6a1a103322901c2603b2","publicPriceSnapshotId":"bigmodel-public-pricing-d5a8f103a28225a9f1e3","internationalPriceSnapshotId":"zai-public-pricing-0f381edee88579cc150d","approvedPlanRulesHash":"057a9ee08689a97d88e70d5ee0ee39c261460f5a57f1e543c65e89b98d4094ec","releaseVersion":"0.21.115","publicationVersion":"277ec5d3b466e1103a93","publicationId":"0097239839f5432f7eda003c"},"asOf":"2026-10-15T11:31:58.983Z","aaSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","sourceSnapshots":{"aa-agent":"aa-agent-7c6f6aac05c8635a05a5","didreset":"didreset-12dea1b25874c069543a","semianalysis-public":"semianalysis-public-c0f1556d65163cf911c6","current-model-products":"current-model-products-2a285087b4ac5668533c","real-api-pricing":"real-api-pricing-da709dedf8368469f99b","opencode-go-terms":"opencode-go-terms-c347cb5d3aa12cc37873","opencode-zen-terms":"opencode-zen-terms-54d03f444331f05ef45c","alibaba-cn-coding":"alibaba-cn-coding-83d936c9575f4123930f","alibaba-int-coding":"alibaba-int-coding-9e93a3c239d884d94351","kimi-membership":"kimi-membership-b5c2f15627657c48b86f","kimi-code-models":"kimi-code-models-beafcace2ba15684a64e","minimax-token-intro":"minimax-token-intro-e91238f24a92ae9de1a1","minimax-token-pricing":"minimax-token-pricing-9bebea789b9a80dfb6df","minimax-cn-terms":"minimax-cn-terms-039399beea272d54abee","grok-build-terms":"grok-build-terms-a086e8833fff246c7740","grok-web-terms":"grok-web-terms-de137bfff621b2d482fa","volcengine-coding-terms":"volcengine-coding-terms-d3b67cbb5f4ffbb267a4","minimax-m-plan-intro":"minimax-m-plan-intro-8e479c68b7f60596aac8","minimax-m-plan-usage":"minimax-m-plan-usage-1b8478dce475f9a34292","minimax-m-plan-pricing":"minimax-m-plan-pricing-43e6f170c77310b5c3e0","minimax-cn-m-plan-intro":"minimax-cn-m-plan-intro-2f6730c07e8953b42a92","minimax-cn-m-plan-usage":"minimax-cn-m-plan-usage-beee2b7b74a9f73eac42","minimax-cn-m-plan-pricing":"minimax-cn-m-plan-pricing-6710562254e049cd42c6","claude-max-api-credits":"claude-max-api-credits-49888e8251805f9903bd","volcengine-personal-coding":"volcengine-personal-coding-068f3e693f3624c7a344","volcengine-enterprise-coding":"volcengine-enterprise-coding-5d7b3c10bbf3b0c7021c","meta-muse-code-subscriptions":"meta-muse-code-subscriptions-976452ec862d2460e246","meta-muse-code-models":"meta-muse-code-models-2079addaef91f7afe932","meta-muse-code-pricing":"meta-muse-code-pricing-6d8285fcb1b1cb234b2a","copilot-model-plan-matrix":"copilot-model-plan-matrix-527f8ef5a2f5f69cc683","copilot-plans":"copilot-plans-6dc752e6c9856326974d","copilot-supported-models":"copilot-supported-models-81cfedb9bd3f5024dc2c","cursor-current-models-pricing":"cursor-current-models-pricing-0765f0d69a40b5362fcd","cursor-current-pricing":"cursor-current-pricing-78b1df5c32d4f0274962","cursor-current-team-pricing":"cursor-current-team-pricing-e584ca0899e06bbe2ca3","kiro-billing":"kiro-billing-23aaffadf9e678b2f758","kiro-models":"kiro-models-9056e99035cb139d24aa","kiro-pricing":"kiro-pricing-5ceb2ec1289d566746db","mimo-model-deprecation":"mimo-model-deprecation-d94e990ae7ad2d5217d6","mimo-personal-subscription":"mimo-personal-subscription-545414c48a5c04f46838","mimo-plan-pricing-faq":"mimo-plan-pricing-faq-70c03a86ac360d538497","mimo-public-team-json":"mimo-public-team-json-8707b9a71604586189f4","mimo-quick-access":"mimo-quick-access-1ca83e1a90a548f6b1bb","mimo-team-subscription":"mimo-team-subscription-567f81812fe752f42bc1","mimo-usage-quota":"mimo-usage-quota-61dacb28640ebbb0a736","mimo-validity-expiry":"mimo-validity-expiry-4943cb1b2bd8f9239876","copilot-marketing":"copilot-marketing-e2c384044993b43293a0","alibaba-cn-token-overview":"alibaba-cn-token-overview-1ffb794900dfb940edf7","alibaba-intl-token-overview":"alibaba-intl-token-overview-47e879dd9e79c313919d","alibaba-cn-token-personal":"alibaba-cn-token-personal-37d52a2f07bb76085ad6","alibaba-intl-token-personal":"alibaba-intl-token-personal-609817a7a93fc6f272fa","alibaba-cn-token-team":"alibaba-cn-token-team-d814b7fa5bf262ebb643","alibaba-intl-token-team":"alibaba-intl-token-team-6fcfe554c68f2e5aec67","volcano-agent-personal-overview":"volcano-agent-personal-overview-20a8671d3ac56d9391a7","volcano-agent-afp-rules":"volcano-agent-afp-rules-c5bab6896a2856c61d16","volcano-agent-small-medium-offer":"volcano-agent-small-medium-offer-a3ad4b26b35122a5c707","volcano-agent-model-offer":"volcano-agent-model-offer-bfbee0a3c719074dfb18","devin-current-pricing":"devin-current-pricing-0b83dc38b00640d4ec52","devin-current-self-serve":"devin-current-self-serve-a2ac32e35538ac7b866f","devin-current-enterprise":"devin-current-enterprise-7388e923e20a5723cad7","devin-current-models":"devin-current-models-bce1288995cc578cba3d","commandcode-current-pricing":"commandcode-current-pricing-da891d6b7debc00e1a8d","commandcode-current-usage-limits":"commandcode-current-usage-limits-e726b6a2c23692de83c9","commandcode-current-go":"commandcode-current-go-bb10ab8c02e0b7d51f9a","commandcode-current-goat":"commandcode-current-goat-701c204b90f305f4093a","commandcode-current-pro":"commandcode-current-pro-10378a82fe3bba8907e8","commandcode-current-max":"commandcode-current-max-f2ced1e8eb4dcf1c1bc3","commandcode-current-faq":"commandcode-current-faq-a643ec6efcca61c3cc5e","commandcode-current-pricing-limits":"commandcode-current-pricing-limits-0fd3bb13cb7beacc8730","google-antigravity-current-pricing":"google-antigravity-current-pricing-d4753b343338852a3c3f","google-ai-current-us-subscriptions":"google-ai-current-us-subscriptions-b5f9afe59bbaae8bb506","google-antigravity-current-models":"google-antigravity-current-models-6aa60f13f860126fba75","google-antigravity-current-plans":"google-antigravity-current-plans-79a99fae66c4b0933d10","google-antigravity-enterprise-conditions":"google-antigravity-enterprise-conditions-8dba2b0cf2711de33726","google-antigravity-plan-price-announcement":"google-antigravity-plan-price-announcement-f120d42d40f2f58953c5","x-subscription-pricing":"x-subscription-pricing-dbba9a70b851444d3630","livebench":"livebench-420a146791c7ae5f33e7","cursorbench":"cursorbench-57c9d63cbcc105c67ec4","coding-model-docs":"coding-model-docs-b25db14ade837bc1f8a5","bigmodel-public-pricing":"bigmodel-public-pricing-d5a8f103a28225a9f1e3","zai-public-pricing":"zai-public-pricing-0f381edee88579cc150d","swebench":"swebench-544c43b6ab8f046b3bde","aider":"aider-dd496c35eb51b52b8bfc","livecodebench":"livecodebench-05e3477cc92d819d65fa","terminalbench":"terminalbench-13678220dcf73850be7e","modelsdev":"modelsdev-604d89749477ce01cca3","openrouter":"openrouter-4fedec27c35d80ccae37","openrouter-popularity":"openrouter-popularity-32976515e400d2923292","openrouter-daily":"openrouter-daily-99c96ede53d13c8dde82","bughunt":"bughunt-3f6a4777d35fbe45d5a6","resets":"resets-6d442b282da4d851ea43","news-models":"news-models-993fd9b2010b1ee0f20d","news-products":"news-products-3bead75ea5a85d53ed69","glm-overview":"glm-overview-58cc06a6f8eb3f64077c","glm-campaign":"glm-campaign-408d970242e101f300c4","glm-legacy":"glm-legacy-3ceef8f37dd55bbd7a22","bigmodel-legacy":"bigmodel-legacy-83b404499e70aa06deb0","bigmodel-overview":"bigmodel-overview-dd848462a1323e1242c2","bigmodel-campaign":"bigmodel-campaign-d3e51f8c01a26b05a32d","openai-pricing":"openai-pricing-0b818c627e5ca186466e","openai-pro-tiers":"openai-pro-tiers-4b12470bb5c07a16e74f","claude-pricing":"claude-pricing-47ef30832dcc59c756dc","zcode-config":"zcode-config-0292ab768aed7a6b1cc9"},"data":[{"id":"sol-5.6","name":"GPT-5.6 Sol","sourceUrl":"https://artificialanalysis.ai/models/releases/gpt-5-6-sol","recommended":"medium","chartReference":{"cost":0.505,"capability":39.2,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":47,"cost":1.9885,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Med，得分 39.2、$0.505/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":47,"boundaryProbePoints":1,"lowestObservedScore":33.5,"relativeFloor":37.6,"headroomFloor":34.85,"effectiveFloor":37.6,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":39.2,"cost":0.505,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"medium","score":39.2,"cost":0.505,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"high","score":42.3,"cost":0.808,"timePenalty":[0,0],"worstRegret":12.311206896551727,"robustWinner":false},{"id":"xhigh","score":44,"cost":1.1844,"timePenalty":[0,0],"worstRegret":29.7556896551724,"robustWinner":false},{"id":"max","score":47,"cost":1.9885,"timePenalty":[0,0],"worstRegret":67.6538793103448,"robustWinner":false}],"robustChoices":["medium"],"boundarySensitive":false,"boundaryProbes":[{"floor":36.6,"winner":"medium"},{"floor":37.6,"winner":"medium"},{"floor":38.6,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium"],"scenarioCount":23,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"none","label":"None","capability":28.3,"cost":null,"evidence":{"sourceId":"7c4d1e30-6ecb-46cf-880c-41446d7b51f1","originalName":"GPT-5.6 Sol (Non-reasoning)","sourceSlug":"gpt-5-6-sol-non-reasoning","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"none","thinkingMode":"disabled","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":null,"method":"unavailable","configurationId":"7c4d1e30-6ecb-46cf-880c-41446d7b51f1","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol-non-reasoning","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:b4eb45efb09a9c4bfdd3e41f4d4059108e92ff008db9b8f8edcd2129bf6b3f16","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["b4eb45efb09a9c4bfdd3e41f4d4059108e92ff008db9b8f8edcd2129bf6b3f16"],"unavailableReason":"no_public_chart_value"},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":33.5,"cost":0.2606,"evidence":{"sourceId":"0904b596-8932-43bd-9b21-324f128e1723","originalName":"GPT-5.6 Sol (Low)","sourceSlug":"gpt-5-6-sol-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":54.81,"method":"aa_published_estimate","configurationId":"0904b596-8932-43bd-9b21-324f128e1723","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol-low","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol-low","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:ae7caaa1e6566002708e2308288656c48cd34a72925647ed101f02a5b0dbdc61","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["ae7caaa1e6566002708e2308288656c48cd34a72925647ed101f02a5b0dbdc61"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":39.2,"cost":0.505,"evidence":{"sourceId":"6f174934-5b7d-4333-86cb-f5ebf4a862e3","originalName":"GPT-5.6 Sol (Medium)","sourceSlug":"gpt-5-6-sol-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":108.16,"method":"aa_published_estimate","configurationId":"6f174934-5b7d-4333-86cb-f5ebf4a862e3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol-medium","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol-medium","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:1bd8db5428bbed23997bfb1bf69d57e70d06104147941a54355e5bd5d61891eb","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["1bd8db5428bbed23997bfb1bf69d57e70d06104147941a54355e5bd5d61891eb"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":42.3,"cost":0.808,"evidence":{"sourceId":"8afc250d-b538-45a2-812a-4605f4ffd87e","originalName":"GPT-5.6 Sol (High)","sourceSlug":"gpt-5-6-sol-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":164.24,"method":"aa_published_estimate","configurationId":"8afc250d-b538-45a2-812a-4605f4ffd87e","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol-high","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol-high","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:7b037a61955e4fbe102595387411876635f812b2a685fed5b627fb151e8aff3f","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["7b037a61955e4fbe102595387411876635f812b2a685fed5b627fb151e8aff3f"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":44,"cost":1.1844,"evidence":{"sourceId":"d998db47-9b67-4727-a2bb-2e1261020ac0","originalName":"GPT-5.6 Sol (Xhigh)","sourceSlug":"gpt-5-6-sol-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":229.56,"method":"aa_published_estimate","configurationId":"d998db47-9b67-4727-a2bb-2e1261020ac0","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol-xhigh","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:98b55a1c00e9a6e31fe8911b871f7d7163e2113fe1fefcd7f472ec2af4644246","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["98b55a1c00e9a6e31fe8911b871f7d7163e2113fe1fefcd7f472ec2af4644246"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":47,"cost":1.9885,"evidence":{"sourceId":"d93edfe8-bf35-49ad-b56e-b18116142a1c","originalName":"GPT-5.6 Sol (Max)","sourceSlug":"gpt-5-6-sol","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":345.12,"method":"aa_published_estimate","configurationId":"d93edfe8-bf35-49ad-b56e-b18116142a1c","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-sol","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-sol","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:7bbd98a506dc5313e7cb5aef4bfbdbe43705fe9774d1c8cabf3a628490cc28aa","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["7bbd98a506dc5313e7cb5aef4bfbdbe43705fe9774d1c8cabf3a628490cc28aa"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:anthropic:2a365f1cb3bef456","name":"Claude Opus 5","sourceUrl":"https://artificialanalysis.ai/models/releases/claude-opus-5-xhigh","recommended":"medium","chartReference":{"cost":2.1895,"capability":44.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":50.8,"cost":5.8584,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Medium，得分 44.8、$2.1895/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":50.8,"boundaryProbePoints":1,"lowestObservedScore":39.4,"relativeFloor":40.64,"headroomFloor":40.54,"effectiveFloor":40.64,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":44.8,"cost":2.1895,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"medium","score":44.8,"cost":2.1895,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"high","score":48.1,"cost":3.6133,"timePenalty":[0,0],"worstRegret":69.11741379310345,"robustWinner":false},{"id":"xhigh","score":49.7,"cost":4.8778,"timePenalty":[0,0],"worstRegret":131.83249999999995,"robustWinner":false},{"id":"max","score":50.8,"cost":5.8584,"timePenalty":[0,0],"worstRegret":180.60784482758615,"robustWinner":false}],"robustChoices":["medium"],"boundarySensitive":false,"boundaryProbes":[{"floor":39.64,"winner":"medium"},{"floor":40.64,"winner":"medium"},{"floor":41.64,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium"],"scenarioCount":23,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"xhigh","label":"Xhigh","capability":49.7,"cost":4.8778,"evidence":{"sourceId":"1305c921-7aaa-4d6d-99b5-99b3acf15e19","originalName":"Claude Opus 5 (Xhigh)","sourceSlug":"claude-opus-5-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":718.02,"method":"aa_published_estimate","configurationId":"1305c921-7aaa-4d6d-99b5-99b3acf15e19","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-xhigh","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-xhigh","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:4e8134756d3bb3a64563cfb5ddcb2a9f0a066929104093378dc536e8820daa7a","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4e8134756d3bb3a64563cfb5ddcb2a9f0a066929104093378dc536e8820daa7a"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":39.4,"cost":1.0983,"evidence":{"sourceId":"20928ba9-3a3f-415f-9519-b84ff64ecf34","originalName":"Claude Opus 5 (Low)","sourceSlug":"claude-opus-5-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":178.22,"method":"aa_published_estimate","configurationId":"20928ba9-3a3f-415f-9519-b84ff64ecf34","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-low","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-low","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:819a1d1af248df190e8663a256e89a38870f92a8198f8b2449346d2e1e3555f6","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["819a1d1af248df190e8663a256e89a38870f92a8198f8b2449346d2e1e3555f6"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":48.1,"cost":3.6133,"evidence":{"sourceId":"712be54a-77ae-41b2-9a58-21181479d6ee","originalName":"Claude Opus 5 (High)","sourceSlug":"claude-opus-5-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":554.51,"method":"aa_published_estimate","configurationId":"712be54a-77ae-41b2-9a58-21181479d6ee","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-high","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-high","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:25d24862b73b9d69d5471270f1ae1d614adf32f64cbaf16c6589c342e1833e21","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["25d24862b73b9d69d5471270f1ae1d614adf32f64cbaf16c6589c342e1833e21"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":50.8,"cost":5.8584,"evidence":{"sourceId":"b8fc61f7-5e9a-49e6-8547-6ac56db24627","originalName":"Claude Opus 5 (Max)","sourceSlug":"claude-opus-5","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":807.81,"method":"aa_published_estimate","configurationId":"b8fc61f7-5e9a-49e6-8547-6ac56db24627","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:ffe435f6725f0860a2b5ee26d95f925b02ea7132f43da9c10db8cb293169dc8a","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["ffe435f6725f0860a2b5ee26d95f925b02ea7132f43da9c10db8cb293169dc8a"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Medium","capability":44.8,"cost":2.1895,"evidence":{"sourceId":"ff51be8f-e362-4a7e-9043-687ba15de207","originalName":"Claude Opus 5 (Medium)","sourceSlug":"claude-opus-5-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":355.6,"method":"aa_published_estimate","configurationId":"ff51be8f-e362-4a7e-9043-687ba15de207","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-medium","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-medium","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:056a3da8ddec1b2531b365586658668c5b03438516660ad0017ab31ea05a5070","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["056a3da8ddec1b2531b365586658668c5b03438516660ad0017ab31ea05a5070"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"astra","name":"GPT-6 Astra","sourceUrl":"https://artificialanalysis.ai/models/releases/gpt-6-astra","recommended":"medium","chartReference":{"cost":1.5406,"capability":49.6,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":52.7,"cost":3.2575,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Med，得分 49.6、$1.5406/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":52.7,"boundaryProbePoints":1,"lowestObservedScore":45.8,"relativeFloor":42.160000000000004,"headroomFloor":46.489999999999995,"effectiveFloor":46.489999999999995,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":49.6,"cost":1.5406,"timePenalty":[0,0],"worstRegret":0.4747637146914627,"robustWinner":false},"rankings":[{"id":"medium","score":49.6,"cost":1.5406,"timePenalty":[0,0],"worstRegret":0.4747637146914627,"robustWinner":false},{"id":"high","score":50.9,"cost":1.7253,"timePenalty":[0,0],"worstRegret":8.09422413793105,"robustWinner":false},{"id":"xhigh","score":52.4,"cost":2.3088,"timePenalty":[0,0],"worstRegret":36.27224137931035,"robustWinner":false},{"id":"max","score":52.7,"cost":3.2575,"timePenalty":[0,0],"worstRegret":84.22508620689655,"robustWinner":false}],"robustChoices":[],"boundarySensitive":true,"boundaryProbes":[{"floor":45.489999999999995,"winner":"low"},{"floor":46.489999999999995,"winner":"medium"},{"floor":47.489999999999995,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["low","medium","high"],"scenarioCount":23,"consistentAcrossScenarios":false,"robustWithinAllRegions":false,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":45.8,"cost":0.8175,"evidence":{"sourceId":"a3f8100d-e38f-408b-b0fa-0085dae18dc1","originalName":"GPT-6 Astra (Low)","sourceSlug":"gpt-6-astra-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":104.95,"method":"aa_published_estimate","configurationId":"a3f8100d-e38f-408b-b0fa-0085dae18dc1","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-astra-low","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-astra-low","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:56092c3fdb953f9c7944c5e8c9f456f8948b0b3ee97d7b854017cdeb2a5d7758","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["56092c3fdb953f9c7944c5e8c9f456f8948b0b3ee97d7b854017cdeb2a5d7758"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":1683.7,"range":[1683.7,1683.7],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"b3ad58f3-b311-4d4d-875b-3158cf0309d6","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Codex","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":50.61,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"medium","label":"Med","capability":49.6,"cost":1.5406,"evidence":{"sourceId":"e97a4ef5-e817-480e-9595-12f81dc4974f","originalName":"GPT-6 Astra (Medium)","sourceSlug":"gpt-6-astra-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":207.92,"method":"aa_published_estimate","configurationId":"e97a4ef5-e817-480e-9595-12f81dc4974f","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-astra-medium","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-astra-medium","sourceObservedAt":"2026-10-14T22:32:30.916Z","sourceEvidenceId":"aa-public-model-page:5e2fc58066dd62b49bde5bda2b727902bbf6c2b64c42d4a8bb2131f550bb3559","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5e2fc58066dd62b49bde5bda2b727902bbf6c2b64c42d4a8bb2131f550bb3559"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":1885.9,"range":[1885.9,1885.9],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"f3c3d5a6-6424-4acb-bfcc-615c3f79f6dd","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Codex","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":54.24,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"high","label":"High","capability":50.9,"cost":1.7253,"evidence":{"sourceId":"e05a4828-0536-4876-870d-a235023f992b","originalName":"GPT-6 Astra (High)","sourceSlug":"gpt-6-astra-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":267.71,"method":"aa_published_estimate","configurationId":"e05a4828-0536-4876-870d-a235023f992b","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-astra-high","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-astra-high","sourceObservedAt":"2026-10-14T22:32:30.916Z","sourceEvidenceId":"aa-public-model-page:5e0ab16d986fe46b037cfde7c28619bec62dae5690283e1fe5d1487dba9d7d2d","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5e0ab16d986fe46b037cfde7c28619bec62dae5690283e1fe5d1487dba9d7d2d"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":2113.7,"range":[2113.7,2113.7],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"3475050c-bf5e-4261-a3f6-0af5350af13f","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Codex","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":57.88,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"xhigh","label":"XHigh","capability":52.4,"cost":2.3088,"evidence":{"sourceId":"1f541ef3-913f-4eb2-9d07-0e93c7a9a5e3","originalName":"GPT-6 Astra (Xhigh)","sourceSlug":"gpt-6-astra-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":387.45,"method":"aa_published_estimate","configurationId":"1f541ef3-913f-4eb2-9d07-0e93c7a9a5e3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-astra-xhigh","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-astra-xhigh","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:6bef1cfa36d59294bbe02e8b9ee2e57610e7c87622517978d9d4757feef54786","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["6bef1cfa36d59294bbe02e8b9ee2e57610e7c87622517978d9d4757feef54786"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":2208.5,"range":[2208.5,2208.5],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"16db8ad5-84aa-4588-b660-1ce68c0d45e2","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Codex","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":57.88,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"max","label":"Max","capability":52.7,"cost":3.2575,"evidence":{"sourceId":"2f339a97-9a0d-499a-9cb5-e0db665bfa25","originalName":"GPT-6 Astra (Max)","sourceSlug":"gpt-6-astra","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":593.42,"method":"aa_published_estimate","configurationId":"2f339a97-9a0d-499a-9cb5-e0db665bfa25","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-astra","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-astra","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:eaadb1a62c88011d94485d029772b211aabe840f0bc7c80d2fda8b4b6f56ca1b","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["eaadb1a62c88011d94485d029772b211aabe840f0bc7c80d2fda8b4b6f56ca1b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":2796.3,"range":[2796.3,2796.3],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"5c537be4-7fc3-449b-8bfc-ceb9061c2535","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Codex","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":58.18,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}}]},{"id":"auto:xai:a58a94661bb3544a","name":"Grok 4.6","sourceUrl":"https://artificialanalysis.ai/models/releases/grok-4-6-medium","recommended":"medium","chartReference":{"cost":1.1438,"capability":42.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"high","capability":44.3,"cost":1.4848,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Medium，得分 42.8、$1.1438/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":44.3,"boundaryProbePoints":1,"lowestObservedScore":35.1,"relativeFloor":35.44,"headroomFloor":36.02,"effectiveFloor":36.02,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":42.8,"cost":1.1438,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"medium","score":42.8,"cost":1.1438,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"high","score":44.3,"cost":1.4848,"timePenalty":[0,0],"worstRegret":15.84396551724138,"robustWinner":false},{"id":"xhigh","score":44.2,"cost":1.7523,"timePenalty":[0,0],"worstRegret":29.54956896551723,"robustWinner":false}],"robustChoices":["medium"],"boundarySensitive":true,"boundaryProbes":[{"floor":35.02,"winner":"low"},{"floor":36.02,"winner":"medium"},{"floor":37.02,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium","low"],"scenarioCount":19,"consistentAcrossScenarios":false,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"medium","label":"Medium","capability":42.8,"cost":1.1438,"evidence":{"sourceId":"26614164-6840-4e17-a65a-2deb2fe7e87b","originalName":"Grok 4.6 (Medium)","sourceSlug":"grok-4-6-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":535.6,"method":"aa_published_estimate","configurationId":"26614164-6840-4e17-a65a-2deb2fe7e87b","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-6-medium","sourceUrl":"https://artificialanalysis.ai/models/grok-4-6-medium","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:04badd95bd63e43cf136965a32256d3ad95435144c79e74631785d5417d76669","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["04badd95bd63e43cf136965a32256d3ad95435144c79e74631785d5417d76669"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":35.1,"cost":0.3758,"evidence":{"sourceId":"c07e65e6-32fc-451e-938d-7477a1c3ffcc","originalName":"Grok 4.6 (Low)","sourceSlug":"grok-4-6-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":186.84,"method":"aa_published_estimate","configurationId":"c07e65e6-32fc-451e-938d-7477a1c3ffcc","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-6-low","sourceUrl":"https://artificialanalysis.ai/models/grok-4-6-low","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:a3cfe6e312a480609b867505220cd66c429ab0ab452289ab6a24eec4915d48a4","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["a3cfe6e312a480609b867505220cd66c429ab0ab452289ab6a24eec4915d48a4"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":44.3,"cost":1.4848,"evidence":{"sourceId":"c8adc5cf-fd5a-407b-af51-dc3bede3e49c","originalName":"Grok 4.6 (High)","sourceSlug":"grok-4-6","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":709.95,"method":"aa_published_estimate","configurationId":"c8adc5cf-fd5a-407b-af51-dc3bede3e49c","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-6","sourceUrl":"https://artificialanalysis.ai/models/grok-4-6","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:cb49602784402cf5c8e036d959dbb1e03c93111c6a56218188f6f14955939c07","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["cb49602784402cf5c8e036d959dbb1e03c93111c6a56218188f6f14955939c07"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"Xhigh","capability":44.2,"cost":1.7523,"evidence":{"sourceId":"d6f43d0a-e8c4-447c-9ff7-5cf0a9d59aa3","originalName":"Grok 4.6 (Xhigh)","sourceSlug":"grok-4-6-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":709.98,"method":"aa_published_estimate","configurationId":"d6f43d0a-e8c4-447c-9ff7-5cf0a9d59aa3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-6-xhigh","sourceUrl":"https://artificialanalysis.ai/models/grok-4-6-xhigh","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:33c13162885956a9f0de2282fdebef0e5cd2da070d868001cc0e9b466a6cf1e7","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["33c13162885956a9f0de2282fdebef0e5cd2da070d868001cc0e9b466a6cf1e7"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"terra-5.6","name":"GPT-5.6 Terra","sourceUrl":"https://artificialanalysis.ai/models/releases/gpt-5-6-terra","recommended":"high","chartReference":{"cost":0.3379,"capability":34.2,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"high","costEffort":"high","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":42.1,"cost":1.3987,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：High，得分 34.2、$0.3379/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":42.1,"boundaryProbePoints":1,"lowestObservedScore":20.8,"relativeFloor":33.68,"headroomFloor":22.93,"effectiveFloor":33.68,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"high","score":34.2,"cost":0.3379,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"high","score":34.2,"cost":0.3379,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"xhigh","score":38,"cost":0.6318,"timePenalty":[0,0],"worstRegret":11.148362068965518,"robustWinner":false},{"id":"max","score":42.1,"cost":1.3987,"timePenalty":[0,0],"worstRegret":46.05448275862068,"robustWinner":false}],"robustChoices":["high"],"boundarySensitive":true,"boundaryProbes":[{"floor":32.68,"winner":"high"},{"floor":33.68,"winner":"high"},{"floor":34.68,"winner":"xhigh"}],"excluded":[{"id":"none","reason":"below-declared-relative-score-protection"},{"id":"low","reason":"below-declared-relative-score-protection"},{"id":"medium","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["high","xhigh"],"scenarioCount":27,"consistentAcrossScenarios":false,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"none","label":"None","capability":20.8,"cost":0.1397,"evidence":{"sourceId":"9b97a35e-6ac7-44d9-91c6-422fa678963e","originalName":"GPT-5.6 Terra (Non-reasoning)","sourceSlug":"gpt-5-6-terra-non-reasoning","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"none","thinkingMode":"disabled","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":34.67,"method":"aa_published_estimate","configurationId":"9b97a35e-6ac7-44d9-91c6-422fa678963e","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra-non-reasoning","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:38b60851337de6a27a2515b0c2c45a21c05f8b973f1df1498944af41902f151b","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["38b60851337de6a27a2515b0c2c45a21c05f8b973f1df1498944af41902f151b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":27.5,"cost":0.1445,"evidence":{"sourceId":"cc4a20cd-09fe-4962-a430-119c815e85fa","originalName":"GPT-5.6 Terra (Low)","sourceSlug":"gpt-5-6-terra-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":39.91,"method":"aa_published_estimate","configurationId":"cc4a20cd-09fe-4962-a430-119c815e85fa","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra-low","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra-low","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:4cd9ff19c3625768a1db145203150af9e89b6fb5fb6032a1e6f40718b44d9578","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4cd9ff19c3625768a1db145203150af9e89b6fb5fb6032a1e6f40718b44d9578"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":30.1,"cost":0.1834,"evidence":{"sourceId":"26e0f83a-ca98-4f34-94ac-7c5e251ee410","originalName":"GPT-5.6 Terra (Medium)","sourceSlug":"gpt-5-6-terra-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":62.8,"method":"aa_published_estimate","configurationId":"26e0f83a-ca98-4f34-94ac-7c5e251ee410","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra-medium","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra-medium","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:5cab31e37ccab25c70ea302b3e66c544268b11953472b1c80e5f91dc779f25ab","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5cab31e37ccab25c70ea302b3e66c544268b11953472b1c80e5f91dc779f25ab"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":34.2,"cost":0.3379,"evidence":{"sourceId":"81972fba-1219-477e-bbfb-18c656a63ff7","originalName":"GPT-5.6 Terra (High)","sourceSlug":"gpt-5-6-terra-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":122.69,"method":"aa_published_estimate","configurationId":"81972fba-1219-477e-bbfb-18c656a63ff7","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra-high","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra-high","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:34b82353acb84f38e5a23fba528a371cd81a9fb6584bd7db9f56fe0fac6e6d70","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["34b82353acb84f38e5a23fba528a371cd81a9fb6584bd7db9f56fe0fac6e6d70"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":38,"cost":0.6318,"evidence":{"sourceId":"9e30696f-16fa-4b4f-ba53-161895a85fed","originalName":"GPT-5.6 Terra (Xhigh)","sourceSlug":"gpt-5-6-terra-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":209.82,"method":"aa_published_estimate","configurationId":"9e30696f-16fa-4b4f-ba53-161895a85fed","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra-xhigh","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:822976688ac916147f7cdc0bc4b393e0c64583564b804c4a6897fb1471b79258","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["822976688ac916147f7cdc0bc4b393e0c64583564b804c4a6897fb1471b79258"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":42.1,"cost":1.3987,"evidence":{"sourceId":"bcf8db0a-3bb6-4d82-9516-0f57370c85a6","originalName":"GPT-5.6 Terra (Max)","sourceSlug":"gpt-5-6-terra","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":340.18,"method":"aa_published_estimate","configurationId":"bcf8db0a-3bb6-4d82-9516-0f57370c85a6","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-5-6-terra","sourceUrl":"https://artificialanalysis.ai/models/gpt-5-6-terra","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:a83779919bd3e731b5f5baa75c8a716b46ee4addb79976015d16b7ebd1a95e44","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["a83779919bd3e731b5f5baa75c8a716b46ee4addb79976015d16b7ebd1a95e44"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:meta:ec63fa545cd577ee","name":"Muse Spark 1.3","sourceUrl":"https://artificialanalysis.ai/models/releases/muse-spark-1-3","recommended":"max","chartReference":{"cost":1.6049,"capability":48.1,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":48.1,"cost":1.6049,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 48.1、$1.6049/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":48.1,"boundaryProbePoints":1,"lowestObservedScore":45.1,"relativeFloor":38.480000000000004,"headroomFloor":45.4,"effectiveFloor":45.4,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":48.1,"cost":1.6049,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"max","score":48.1,"cost":1.6049,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["max"],"boundarySensitive":true,"boundaryProbes":[{"floor":44.4,"winner":"xhigh"},{"floor":45.4,"winner":"max"},{"floor":46.4,"winner":"max"}],"excluded":[{"id":"xhigh","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["max","xhigh"],"scenarioCount":11,"consistentAcrossScenarios":false,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"max","label":"Max","capability":48.1,"cost":1.6049,"evidence":{"sourceId":"9999672c-b687-4026-9c15-5cef02ff53bb","originalName":"Muse Spark 1.3 (Max)","sourceSlug":"muse-spark-1-3","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":464.41,"method":"aa_published_estimate","configurationId":"9999672c-b687-4026-9c15-5cef02ff53bb","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"muse-spark-1-3","sourceUrl":"https://artificialanalysis.ai/models/muse-spark-1-3","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:4b54da8371c1238f8a82b161c385bf84e16fd298bb317f2e284cde94e4d90a30","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4b54da8371c1238f8a82b161c385bf84e16fd298bb317f2e284cde94e4d90a30"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"Xhigh","capability":45.1,"cost":1.3678,"evidence":{"sourceId":"d5170215-69be-4129-849b-26d8d8825bfc","originalName":"Muse Spark 1.3 (Xhigh)","sourceSlug":"muse-spark-1-3-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":202.76,"method":"aa_published_estimate","configurationId":"d5170215-69be-4129-849b-26d8d8825bfc","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"muse-spark-1-3-xhigh","sourceUrl":"https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:eba116c95adf861af2277e49581f1d4f2a357970a5fdf081431e80744895877f","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["eba116c95adf861af2277e49581f1d4f2a357970a5fdf081431e80744895877f"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:alibaba:b6b89c36a7e23db5","name":"Qwen3.8 Max (0902)","sourceUrl":"https://artificialanalysis.ai/models/releases/qwen3-8-max","recommended":"observed","chartReference":{"cost":5.4085,"capability":45.4,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":45.4,"cost":5.4085,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":45.4,"cost":5.4085,"evidence":{"sourceId":"b112af07-3bd5-4647-b09f-b23204361bb2","originalName":"Qwen3.8 Max (0902)","sourceSlug":"qwen3-8-max","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":2160.54,"method":"aa_published_estimate","configurationId":"b112af07-3bd5-4647-b09f-b23204361bb2","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"qwen3-8-max","sourceUrl":"https://artificialanalysis.ai/models/qwen3-8-max","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:6b2792d79d3b2ce09b2894f9b9577be4c0f64e0860cbde9021c31416d7f06807","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["6b2792d79d3b2ce09b2894f9b9577be4c0f64e0860cbde9021c31416d7f06807"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:alibaba:d3a69ebbb2a99c05","name":"Qwen3.8-Flash-Next","sourceUrl":"https://artificialanalysis.ai/models/releases/qwen3-8-flash-next","recommended":"observed","chartReference":{"cost":0.3722,"capability":39.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":39.8,"cost":0.3722,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":39.8,"cost":0.3722,"evidence":{"sourceId":"9ac3908c-9eb2-44db-8c01-3ec7da6fc2ae","originalName":"Qwen3.8-Flash-Next","sourceSlug":"qwen3-8-flash-next","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":1666.45,"method":"aa_published_estimate","configurationId":"9ac3908c-9eb2-44db-8c01-3ec7da6fc2ae","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"qwen3-8-flash-next","sourceUrl":"https://artificialanalysis.ai/models/qwen3-8-flash-next","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:e546e44559b7214bbbd7868ff0e7868a08e2793567493af9bff0fdf0b682fa88","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["e546e44559b7214bbbd7868ff0e7868a08e2793567493af9bff0fdf0b682fa88"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"glm","name":"GLM-5.3","sourceUrl":"https://artificialanalysis.ai/models/glm-5-3","recommended":"max","chartReference":{"cost":2.0056,"capability":44.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":44.8,"cost":2.0056,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 44.8、$2.0056/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":44.8,"boundaryProbePoints":1,"lowestObservedScore":34.3,"relativeFloor":35.839999999999996,"headroomFloor":35.349999999999994,"effectiveFloor":35.839999999999996,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":44.8,"cost":2.0056,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"max","score":44.8,"cost":2.0056,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["max"],"boundarySensitive":false,"boundaryProbes":[{"floor":34.839999999999996,"winner":"max"},{"floor":35.839999999999996,"winner":"max"},{"floor":36.839999999999996,"winner":"max"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["max"],"scenarioCount":11,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":34.3,"cost":0.8522,"evidence":{"sourceId":"cc1e488b-8ba4-464d-875d-690166b15ad3","originalName":"GLM-5.3 (Low)","sourceSlug":"glm-5-3-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":445.34,"method":"aa_published_estimate","configurationId":"cc1e488b-8ba4-464d-875d-690166b15ad3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"glm-5-3-low","sourceUrl":"https://artificialanalysis.ai/models/glm-5-3-low","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:26f7a94e1d93c34ea13097d11212327aaecdd8e74ab42a594f451d660dc48475","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["26f7a94e1d93c34ea13097d11212327aaecdd8e74ab42a594f451d660dc48475"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":44.8,"cost":2.0056,"evidence":{"sourceId":"cd684ea4-b475-4269-b001-d469d06d8a7a","originalName":"GLM-5.3 (Max)","sourceSlug":"glm-5-3","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":775.59,"method":"aa_published_estimate","configurationId":"cd684ea4-b475-4269-b001-d469d06d8a7a","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"glm-5-3","sourceUrl":"https://artificialanalysis.ai/models/glm-5-3","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:5d24c61d95c81dadc2d7894bf61b5007ef6c832b00ffac43a8d8a78a1fef19e7","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5d24c61d95c81dadc2d7894bf61b5007ef6c832b00ffac43a8d8a78a1fef19e7"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":5831,"range":[5831,5831],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"d72e8775-f4f1-4313-99e7-35b1cb499f24","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":41.82,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}}]},{"id":"opus","name":"Claude Opus 5.5","sourceUrl":"https://artificialanalysis.ai/models/releases/claude-opus-5-5","recommended":"medium","chartReference":{"cost":1.336,"capability":51.2,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":57.6,"cost":5.982,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Med，得分 51.2、$1.336/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":57.6,"boundaryProbePoints":1,"lowestObservedScore":42.3,"relativeFloor":46.080000000000005,"headroomFloor":43.83,"effectiveFloor":46.080000000000005,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":51.2,"cost":1.336,"timePenalty":[0,0],"worstRegret":0.13679980598242736,"robustWinner":false},"rankings":[{"id":"medium","score":51.2,"cost":1.336,"timePenalty":[0,0],"worstRegret":0.13679980598242736,"robustWinner":false},{"id":"high","score":53.6,"cost":1.8225,"timePenalty":[0,0],"worstRegret":22.344396551724138,"robustWinner":false},{"id":"xhigh","score":56,"cost":3.4591,"timePenalty":[0,0],"worstRegret":103.18525862068962,"robustWinner":false},{"id":"max","score":57.6,"cost":5.982,"timePenalty":[0,0],"worstRegret":229.90517241379308,"robustWinner":false}],"robustChoices":[],"boundarySensitive":false,"boundaryProbes":[{"floor":45.080000000000005,"winner":"medium"},{"floor":46.080000000000005,"winner":"medium"},{"floor":47.080000000000005,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium"],"scenarioCount":23,"consistentAcrossScenarios":true,"robustWithinAllRegions":false,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":42.3,"cost":0.5512,"evidence":{"sourceId":"4416ca93-fde6-4b7b-abbc-3a6be109cc3d","originalName":"Claude Opus 5.5 (Low, Default Fallback)","sourceSlug":"claude-opus-5-5-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":86.63,"method":"aa_published_estimate","configurationId":"4416ca93-fde6-4b7b-abbc-3a6be109cc3d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-5-low","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-5-low","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:d6b32f612794a8191bde1cbbf9d86a87face4cbc2af142a2e7f907f3c2c6271b","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["d6b32f612794a8191bde1cbbf9d86a87face4cbc2af142a2e7f907f3c2c6271b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":51.2,"cost":1.336,"evidence":{"sourceId":"ea9e0da6-169f-43c6-92f5-730a10d6ff8d","originalName":"Claude Opus 5.5 (Medium, Default Fallback)","sourceSlug":"claude-opus-5-5-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":219.84,"method":"aa_published_estimate","configurationId":"ea9e0da6-169f-43c6-92f5-730a10d6ff8d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-5-medium","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-5-medium","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:4207c7b39b41ec862f1757aaa8fd1f09cf304072017cfce0e9c950d3fa43d362","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4207c7b39b41ec862f1757aaa8fd1f09cf304072017cfce0e9c950d3fa43d362"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":53.6,"cost":1.8225,"evidence":{"sourceId":"f314eade-2232-44bf-a3ab-9b326bc67de6","originalName":"Claude Opus 5.5 (High, Default Fallback)","sourceSlug":"claude-opus-5-5-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":291.93,"method":"aa_published_estimate","configurationId":"f314eade-2232-44bf-a3ab-9b326bc67de6","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-5-high","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-5-high","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:275b533f731c4824655c842f8e9b613d43bf1252fd0b6f9d520eafbf4d53ff0c","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["275b533f731c4824655c842f8e9b613d43bf1252fd0b6f9d520eafbf4d53ff0c"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":56,"cost":3.4591,"evidence":{"sourceId":"df5bf99d-d228-4715-b41b-19ca86118777","originalName":"Claude Opus 5.5 (Xhigh, Default Fallback)","sourceSlug":"claude-opus-5-5-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":519.73,"method":"aa_published_estimate","configurationId":"df5bf99d-d228-4715-b41b-19ca86118777","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-5-xhigh","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:b46cc548b2016f9e03385f9bfbb3426c5349268d26f0d882bfa2d83181a11d32","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["b46cc548b2016f9e03385f9bfbb3426c5349268d26f0d882bfa2d83181a11d32"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":57.6,"cost":5.982,"evidence":{"sourceId":"2f3c4dc9-a450-4303-8697-0237257cf08f","originalName":"Claude Opus 5.5 (Max, Default Fallback)","sourceSlug":"claude-opus-5-5","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":797.48,"method":"aa_published_estimate","configurationId":"2f3c4dc9-a450-4303-8697-0237257cf08f","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-opus-5-5","sourceUrl":"https://artificialanalysis.ai/models/claude-opus-5-5","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:0afe69d4e2c4dc226507ac52b22a88751dfed56d95187c5bbd3f98b472b8e10e","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["0afe69d4e2c4dc226507ac52b22a88751dfed56d95187c5bbd3f98b472b8e10e"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":{"kind":"historical-reference","seconds":3960}}}]},{"id":"auto:google:f83b6872695c3bdd","name":"Gemini 4 Argon","sourceUrl":"https://artificialanalysis.ai/models/releases/gemini-4-argon","recommended":"high","chartReference":{"cost":1.9903,"capability":52.6,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"high","costEffort":"high","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"high","capability":52.6,"cost":1.9903,"benchmark":"AA Intelligence Index"},"recommendationReason":"只有 High 一组评测配置，推荐档就是这一档；没有其他档位可比。","recommendationPolicy":{"basis":"single-configuration","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"high","label":"High","capability":52.6,"cost":1.9903,"evidence":{"sourceId":"d366dfa1-d940-4788-aa73-d298c45699b3","originalName":"Gemini 4 Argon (High)","sourceSlug":"gemini-4-argon","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":null,"method":"unavailable","configurationId":"d366dfa1-d940-4788-aa73-d298c45699b3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gemini-4-argon","sourceUrl":"https://artificialanalysis.ai/models/gemini-4-argon","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:7da4dc744b12ec44a070ec1d345fa8638470c3ac5ff9232c8d4755e446f433e1","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["7da4dc744b12ec44a070ec1d345fa8638470c3ac5ff9232c8d4755e446f433e1"],"unavailableReason":"no_public_chart_value"},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:meta:beb11fe7850d6b1b","name":"Muse Wonder","sourceUrl":"https://artificialanalysis.ai/models/releases/muse-wonder","recommended":"max","chartReference":{"cost":4.5157,"capability":52.4,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":52.4,"cost":4.5157,"benchmark":"AA Intelligence Index"},"recommendationReason":"只有 Max 一组评测配置，推荐档就是这一档；没有其他档位可比。","recommendationPolicy":{"basis":"single-configuration","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"max","label":"Max","capability":52.4,"cost":4.5157,"evidence":{"sourceId":"ff2f3f3d-783f-4e27-8a39-47237501a313","originalName":"Muse Wonder (Max)","sourceSlug":"muse-wonder","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":409.51,"method":"aa_published_estimate","configurationId":"ff2f3f3d-783f-4e27-8a39-47237501a313","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"muse-wonder","sourceUrl":"https://artificialanalysis.ai/models/muse-wonder","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:5479a926e1beed95c84648e3620d4a967cf875f716b44cb921ee96932daa43a7","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5479a926e1beed95c84648e3620d4a967cf875f716b44cb921ee96932daa43a7"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"glm-5.3-flash","name":"GLM-5.3-Flash","sourceUrl":"https://artificialanalysis.ai/models/releases/glm-5-3-flash","recommended":"max","chartReference":{"cost":0.2836,"capability":41.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":41.8,"cost":0.2836,"benchmark":"AA Intelligence Index"},"recommendationReason":"只有 Max 一组评测配置，推荐档就是这一档；没有其他档位可比。","recommendationPolicy":{"basis":"single-configuration","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"max","label":"Max","capability":41.8,"cost":0.2836,"evidence":{"sourceId":"19496b81-9f41-4214-a77a-1df803b3c5ae","originalName":"GLM-5.3-Flash","sourceSlug":"glm-5-3-flash","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":768.49,"method":"aa_published_estimate","configurationId":"19496b81-9f41-4214-a77a-1df803b3c5ae","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"glm-5-3-flash","sourceUrl":"https://artificialanalysis.ai/models/glm-5-3-flash","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:22cef806ab2d5539233c289635720cdb0d69f7a01d59305c29c6ed4874246031","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["22cef806ab2d5539233c289635720cdb0d69f7a01d59305c29c6ed4874246031"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"sol-6.1","name":"GPT-6.1 Sol","sourceUrl":"https://artificialanalysis.ai/models/releases/gpt-6-1-sol","recommended":"medium","chartReference":{"cost":0.2137,"capability":47.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":51.8,"cost":0.7242,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Med，得分 47.8、$0.2137/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":51.8,"boundaryProbePoints":1,"lowestObservedScore":42.1,"relativeFloor":41.44,"headroomFloor":43.07,"effectiveFloor":43.07,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":47.8,"cost":0.2137,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"medium","score":47.8,"cost":0.2137,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"high","score":50.2,"cost":0.3191,"timePenalty":[0,0],"worstRegret":2.960862068965504,"robustWinner":false},{"id":"xhigh","score":51,"cost":0.3929,"timePenalty":[0,0],"worstRegret":5.914482758620686,"robustWinner":false},{"id":"max","score":51.8,"cost":0.7242,"timePenalty":[0,0],"worstRegret":21.965086206896544,"robustWinner":false}],"robustChoices":["medium"],"boundarySensitive":false,"boundaryProbes":[{"floor":42.07,"winner":"medium"},{"floor":43.07,"winner":"medium"},{"floor":44.07,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium"],"scenarioCount":23,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":42.1,"cost":0.1308,"evidence":{"sourceId":"e76e0fe5-93ca-4272-9359-2eb1740467d1","originalName":"GPT-6.1 Sol (Low)","sourceSlug":"gpt-6-1-sol-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":88.73,"method":"aa_published_estimate","configurationId":"e76e0fe5-93ca-4272-9359-2eb1740467d1","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-1-sol-low","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-1-sol-low","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:234d00b0c1edc4b3b8fecfed24f8c050f01c24f1fd9861b24c25488f26ce858a","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["234d00b0c1edc4b3b8fecfed24f8c050f01c24f1fd9861b24c25488f26ce858a"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":47.8,"cost":0.2137,"evidence":{"sourceId":"a2e51d62-ea4e-4f15-8be0-be9dc1fb489d","originalName":"GPT-6.1 Sol (Medium)","sourceSlug":"gpt-6-1-sol-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":178.31,"method":"aa_published_estimate","configurationId":"a2e51d62-ea4e-4f15-8be0-be9dc1fb489d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-1-sol-medium","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-1-sol-medium","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:48900b1be78b1ff6faf0676670ea4db799b3cd515f9b0d85e2ce0f7460b8f6f4","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["48900b1be78b1ff6faf0676670ea4db799b3cd515f9b0d85e2ce0f7460b8f6f4"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":50.2,"cost":0.3191,"evidence":{"sourceId":"3b84fee3-b70a-41bc-819e-5fbdef9a0042","originalName":"GPT-6.1 Sol (High)","sourceSlug":"gpt-6-1-sol-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":267.48,"method":"aa_published_estimate","configurationId":"3b84fee3-b70a-41bc-819e-5fbdef9a0042","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-1-sol-high","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-1-sol-high","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:611dbcff8c4537fd62ed3b6f1cec954af08cb8003230be9d0a5ae257aa8fe04b","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["611dbcff8c4537fd62ed3b6f1cec954af08cb8003230be9d0a5ae257aa8fe04b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":51,"cost":0.3929,"evidence":{"sourceId":"092a3b0e-c5c8-45dc-bf1b-53673c8ff352","originalName":"GPT-6.1 Sol (Xhigh)","sourceSlug":"gpt-6-1-sol-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":341.91,"method":"aa_published_estimate","configurationId":"092a3b0e-c5c8-45dc-bf1b-53673c8ff352","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-1-sol-xhigh","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:8df1f8bf1ad124a3cce19d0a8eef5fbc828226f6d429199d874b27aadc1fb3b5","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["8df1f8bf1ad124a3cce19d0a8eef5fbc828226f6d429199d874b27aadc1fb3b5"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":51.8,"cost":0.7242,"evidence":{"sourceId":"b25ac058-1e33-4c31-bbc9-a20c3ad0717a","originalName":"GPT-6.1 Sol (Max)","sourceSlug":"gpt-6-1-sol","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":797.27,"method":"aa_published_estimate","configurationId":"b25ac058-1e33-4c31-bbc9-a20c3ad0717a","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-1-sol","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-1-sol","sourceObservedAt":"2026-10-15T00:47:07.239Z","sourceEvidenceId":"aa-public-model-page:64051da1ac1cbbcddd403205b7e7e0360286bcdfc906a583e6eb3f3b78721b41","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["64051da1ac1cbbcddd403205b7e7e0360286bcdfc906a583e6eb3f3b78721b41"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:google:0bb591b009df2469","name":"Gemini 3.8 Flash","sourceUrl":"https://artificialanalysis.ai/models/releases/gemini-3-8-flash-medium","recommended":"high","chartReference":{"cost":1.2428,"capability":40.9,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"high","costEffort":"high","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"high","capability":40.9,"cost":1.2428,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：High，得分 40.9、$1.2428/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":40.9,"boundaryProbePoints":1,"lowestObservedScore":39.8,"relativeFloor":32.72,"headroomFloor":39.91,"effectiveFloor":39.91,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"high","score":40.9,"cost":1.2428,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"high","score":40.9,"cost":1.2428,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["high"],"boundarySensitive":true,"boundaryProbes":[{"floor":38.91,"winner":"medium"},{"floor":39.91,"winner":"high"},{"floor":40.91,"winner":null}],"excluded":[{"id":"medium","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["medium","high"],"scenarioCount":11,"consistentAcrossScenarios":false,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"medium","label":"Medium","capability":39.8,"cost":0.931,"evidence":{"sourceId":"744fa3c1-42a7-4995-b138-fb6be11f463c","originalName":"Gemini 3.8 Flash (Medium)","sourceSlug":"gemini-3-8-flash-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":null,"method":"unavailable","configurationId":"744fa3c1-42a7-4995-b138-fb6be11f463c","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gemini-3-8-flash-medium","sourceUrl":"https://artificialanalysis.ai/models/gemini-3-8-flash-medium","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:4c5102f688f9cfc61b551bda41eecb226923b48cb86a51af1e84388554798a0c","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4c5102f688f9cfc61b551bda41eecb226923b48cb86a51af1e84388554798a0c"],"unavailableReason":"no_public_chart_value"},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":40.9,"cost":1.2428,"evidence":{"sourceId":"8a999846-4c1d-4ce7-a8b7-1310a7166fd7","originalName":"Gemini 3.8 Flash (High)","sourceSlug":"gemini-3-8-flash","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":494.59,"method":"aa_published_estimate","configurationId":"8a999846-4c1d-4ce7-a8b7-1310a7166fd7","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gemini-3-8-flash","sourceUrl":"https://artificialanalysis.ai/models/gemini-3-8-flash","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:74a4dbf757b6ba459f0a6fae5591ca37f54a8a114ddcfe42dab47b28e0254100","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["74a4dbf757b6ba459f0a6fae5591ca37f54a8a114ddcfe42dab47b28e0254100"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":33.5,"cost":null,"evidence":{"sourceId":"f6db039b-0f3b-485c-9d7f-982988e44f26","originalName":"Gemini 3.8 Flash (Low)","sourceSlug":"gemini-3-8-flash-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":null,"method":"unavailable","configurationId":"f6db039b-0f3b-485c-9d7f-982988e44f26","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gemini-3-8-flash-low","sourceUrl":"https://artificialanalysis.ai/models/gemini-3-8-flash-low","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:3012e57ca29648d787d5dff300127a50f581129c06b041cfb69f0ca3d0fef18a","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["3012e57ca29648d787d5dff300127a50f581129c06b041cfb69f0ca3d0fef18a"],"unavailableReason":"no_public_chart_value"},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:kimi:d59a3fc7d1ed2d19","name":"Kimi K3","sourceUrl":"https://artificialanalysis.ai/models/releases/kimi-k3-low","recommended":"max","chartReference":{"cost":2.0001,"capability":43.6,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":43.6,"cost":2.0001,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 43.6、$2.0001/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":43.6,"boundaryProbePoints":1,"lowestObservedScore":30.1,"relativeFloor":34.88,"headroomFloor":31.450000000000003,"effectiveFloor":34.88,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":43.6,"cost":2.0001,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"max","score":43.6,"cost":2.0001,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["max"],"boundarySensitive":false,"boundaryProbes":[{"floor":33.88,"winner":"max"},{"floor":34.88,"winner":"max"},{"floor":35.88,"winner":"max"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["max"],"scenarioCount":11,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":30.1,"cost":1.1463,"evidence":{"sourceId":"512d17ef-13d2-4f65-bf9b-154b0dec7e8d","originalName":"Kimi K3 (Low)","sourceSlug":"kimi-k3-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":283.26,"method":"aa_published_estimate","configurationId":"512d17ef-13d2-4f65-bf9b-154b0dec7e8d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"kimi-k3-low","sourceUrl":"https://artificialanalysis.ai/models/kimi-k3-low","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:5c2e1f41fbc27db9a3c9c83ee21eaf0249749801735fac2f1241aa459eb02500","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["5c2e1f41fbc27db9a3c9c83ee21eaf0249749801735fac2f1241aa459eb02500"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":43.6,"cost":2.0001,"evidence":{"sourceId":"f7d2fc3e-1f7b-405f-818c-07952a4af78f","originalName":"Kimi K3 (Max)","sourceSlug":"kimi-k3","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":989.6,"method":"aa_published_estimate","configurationId":"f7d2fc3e-1f7b-405f-818c-07952a4af78f","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"kimi-k3","sourceUrl":"https://artificialanalysis.ai/models/kimi-k3","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:143e720fa061876fe808967fa5ce9f089642d7a5972713e61c7aa1c5210cbdb1","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["143e720fa061876fe808967fa5ce9f089642d7a5972713e61c7aa1c5210cbdb1"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"fable","name":"Claude Fable 5.1","sourceUrl":"https://artificialanalysis.ai/models/releases/claude-fable-5-1","recommended":"medium","chartReference":{"cost":2.9826,"capability":48.9,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"medium","costEffort":"medium","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":53.4,"cost":7.6297,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Med，得分 48.9、$2.9826/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":53.4,"boundaryProbePoints":1,"lowestObservedScore":46.8,"relativeFloor":42.72,"headroomFloor":47.459999999999994,"effectiveFloor":47.459999999999994,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"medium","score":48.9,"cost":2.9826,"timePenalty":[0,0],"worstRegret":0.3221181922101195,"robustWinner":false},"rankings":[{"id":"medium","score":48.9,"cost":2.9826,"timePenalty":[0,0],"worstRegret":0.3221181922101195,"robustWinner":false},{"id":"high","score":51.2,"cost":3.9125,"timePenalty":[0,0],"worstRegret":44.996637931034485,"robustWinner":false},{"id":"xhigh","score":53.2,"cost":5.9783,"timePenalty":[0,0],"worstRegret":148.06749999999997,"robustWinner":false},{"id":"max","score":53.4,"cost":7.6297,"timePenalty":[0,0],"worstRegret":231.8611206896551,"robustWinner":false}],"robustChoices":[],"boundarySensitive":true,"boundaryProbes":[{"floor":46.459999999999994,"winner":"low"},{"floor":47.459999999999994,"winner":"medium"},{"floor":48.459999999999994,"winner":"medium"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["low","medium"],"scenarioCount":23,"consistentAcrossScenarios":false,"robustWithinAllRegions":false,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":46.8,"cost":2.371,"evidence":{"sourceId":"05776db7-f5c0-40f7-b824-079160f8cfa8","originalName":"Claude Fable 5.1 (Low, Default Fallback)","sourceSlug":"claude-fable-5-1-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":261.1,"method":"aa_published_estimate","configurationId":"05776db7-f5c0-40f7-b824-079160f8cfa8","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-fable-5-1-low","sourceUrl":"https://artificialanalysis.ai/models/claude-fable-5-1-low","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:b7a06c0f0198d68f223e8a39e50a3d79bfa5266a70889f5b0331c68a2d1c4a74","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["b7a06c0f0198d68f223e8a39e50a3d79bfa5266a70889f5b0331c68a2d1c4a74"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":2290.8,"range":[2290.8,2290.8],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"e18d02ab-0973-4d86-872f-a353c8a7f344","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":43.33,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"medium","label":"Med","capability":48.9,"cost":2.9826,"evidence":{"sourceId":"3b7de71c-e034-4591-8ca6-6b6be2fa471f","originalName":"Claude Fable 5.1 (Medium, Default Fallback)","sourceSlug":"claude-fable-5-1-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":341.41,"method":"aa_published_estimate","configurationId":"3b7de71c-e034-4591-8ca6-6b6be2fa471f","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-fable-5-1-medium","sourceUrl":"https://artificialanalysis.ai/models/claude-fable-5-1-medium","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:57b21fdb58f65a9a1e5ed09659943ec7583199c907c511e0d4c1fc9324812acb","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["57b21fdb58f65a9a1e5ed09659943ec7583199c907c511e0d4c1fc9324812acb"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":2528.3,"range":[2528.3,2528.3],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"83e6bf7e-aeb4-4116-9f30-ca8de96603fb","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":53.94,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"high","label":"High","capability":51.2,"cost":3.9125,"evidence":{"sourceId":"9d7d72cd-d95d-45a0-b109-4ad292c9aabd","originalName":"Claude Fable 5.1 (High, Default Fallback)","sourceSlug":"claude-fable-5-1-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":450.21,"method":"aa_published_estimate","configurationId":"9d7d72cd-d95d-45a0-b109-4ad292c9aabd","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-fable-5-1-high","sourceUrl":"https://artificialanalysis.ai/models/claude-fable-5-1-high","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:7710646745862f208887d698b99f3b10226116912ec765d0d6f3452b576f604e","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["7710646745862f208887d698b99f3b10226116912ec765d0d6f3452b576f604e"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":3017.7,"range":[3017.7,3017.7],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"4643b23c-53e8-47ae-a3bd-9abc900e9588","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":54.55,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"xhigh","label":"XHigh","capability":53.2,"cost":5.9783,"evidence":{"sourceId":"9b166bf3-42db-4f63-8338-1c4a1244ffe8","originalName":"Claude Fable 5.1 (Xhigh, Default Fallback)","sourceSlug":"claude-fable-5-1-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":608.06,"method":"aa_published_estimate","configurationId":"9b166bf3-42db-4f63-8338-1c4a1244ffe8","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-fable-5-1-xhigh","sourceUrl":"https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:3b1a8348502e776ea73f322c599c65d9b160616659799a28038a2cad3d9e583a","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["3b1a8348502e776ea73f322c599c65d9b160616659799a28038a2cad3d9e583a"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":3202.1,"range":[3202.1,3202.1],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"9e1c5e11-748f-4f2e-911a-070a396d5bf8","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":57.88,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}},{"effort":"max","label":"Max","capability":53.4,"cost":7.6297,"evidence":{"sourceId":"3e87c73e-a257-495e-9730-367a66229811","originalName":"Claude Fable 5.1 (Max, Default Fallback)","sourceSlug":"claude-fable-5-1","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":687.43,"method":"aa_published_estimate","configurationId":"3e87c73e-a257-495e-9730-367a66229811","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-fable-5-1","sourceUrl":"https://artificialanalysis.ai/models/claude-fable-5-1","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:0c6b0faead2531c5847da73315be72b2c5c339fabea4c5c31c4e6ded23b0ca0a","fallbackPolicy":"Default Fallback","excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["0c6b0faead2531c5847da73315be72b2c5c339fabea4c5c31c4e6ded23b0ca0a"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"measured","seconds":3893.5,"range":[3893.5,3893.5],"rankingEligible":true,"evidence":{"source":"Terminal-Bench 4.0","recordId":"c741608e-c94e-417d-b7a8-e67111a0c887","sourceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/4.0","snapshotId":"terminalbench-13678220dcf73850be7e","scope":"trial_reported","harness":"Claude Code","harnessVersion":null,"fallbackPolicy":null,"protocolId":"terminal-bench-4-0-0:1922072f-a433-429a-8929-350d5e1bcf02","taskSetId":"1922072f-a433-429a-8929-350d5e1bcf02","timingBoundary":"trial_wall","statistic":"mean","population":"attempts-with-parseable-start-and-finish","reportedTrials":330,"timedTrials":null,"successRatePercent":57.88,"aggregationCodeUrl":"https://github.com/harbor-framework/terminal-bench/blob/4def1f367467b34b18e0dbdc086400ba71c3e037/leaderboard/src/leaderboard/core/metrics.py","observedAt":"2026-10-07T01:46:30.385Z","stale":false}}}]},{"id":"auto:xai:208666a1a50e881a","name":"Grok 4.7","sourceUrl":"https://artificialanalysis.ai/models/releases/grok-4-7-low","recommended":"high","chartReference":{"cost":2.7261,"capability":46.3,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"high","costEffort":"high","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"xhigh","capability":46.4,"cost":3.7383,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：High，得分 46.3、$2.7261/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":46.4,"boundaryProbePoints":1,"lowestObservedScore":42.2,"relativeFloor":37.12,"headroomFloor":42.620000000000005,"effectiveFloor":42.620000000000005,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"high","score":46.3,"cost":2.7261,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"high","score":46.3,"cost":2.7261,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"xhigh","score":46.4,"cost":3.7383,"timePenalty":[0,0],"worstRegret":51.382586206896534,"robustWinner":false}],"robustChoices":["high"],"boundarySensitive":true,"boundaryProbes":[{"floor":41.620000000000005,"winner":"low"},{"floor":42.620000000000005,"winner":"high"},{"floor":43.620000000000005,"winner":"high"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["low","high"],"scenarioCount":15,"consistentAcrossScenarios":false,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":42.2,"cost":1.248,"evidence":{"sourceId":"208c16f0-8c5b-4524-ac8a-ac74c02e80a9","originalName":"Grok 4.7 (Low)","sourceSlug":"grok-4-7-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":461.31,"method":"aa_published_estimate","configurationId":"208c16f0-8c5b-4524-ac8a-ac74c02e80a9","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-7-low","sourceUrl":"https://artificialanalysis.ai/models/grok-4-7-low","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:f8d8b9de90fd9ffe3abe69281b40ec4628609978f943a623de6396ded87127de","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["f8d8b9de90fd9ffe3abe69281b40ec4628609978f943a623de6396ded87127de"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":46.3,"cost":2.7261,"evidence":{"sourceId":"272f5f03-aea9-4675-a244-2b753b618cdf","originalName":"Grok 4.7 (High)","sourceSlug":"grok-4-7-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":938.08,"method":"aa_published_estimate","configurationId":"272f5f03-aea9-4675-a244-2b753b618cdf","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-7-high","sourceUrl":"https://artificialanalysis.ai/models/grok-4-7-high","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:572ef0868cb3084cab60c09a555b8f23c0a477ad31d1f4d75c680b2fe92ffa43","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["572ef0868cb3084cab60c09a555b8f23c0a477ad31d1f4d75c680b2fe92ffa43"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"Xhigh","capability":46.4,"cost":3.7383,"evidence":{"sourceId":"59627802-1546-422c-9a65-65b61c022d36","originalName":"Grok 4.7 (Xhigh)","sourceSlug":"grok-4-7","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":1022.31,"method":"aa_published_estimate","configurationId":"59627802-1546-422c-9a65-65b61c022d36","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"grok-4-7","sourceUrl":"https://artificialanalysis.ai/models/grok-4-7","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:892c6de4b193ce41232f897bcdc97ded5c72fcc80ba44f6b1e6d7af50b26b729","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["892c6de4b193ce41232f897bcdc97ded5c72fcc80ba44f6b1e6d7af50b26b729"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:xiaomi:5606e136097c03fb","name":"MiMo-V2.6-Pro","sourceUrl":"https://artificialanalysis.ai/models/releases/mimo-v2-6-pro","recommended":"observed","chartReference":{"cost":0.1332,"capability":46.3,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":46.3,"cost":0.1332,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":46.3,"cost":0.1332,"evidence":{"sourceId":"24d8fdfe-6241-424e-b7b7-1b6efb06e4fb","originalName":"MiMo-V2.6-Pro","sourceSlug":"mimo-v2-6-pro","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":1346.05,"method":"aa_published_estimate","configurationId":"24d8fdfe-6241-424e-b7b7-1b6efb06e4fb","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"mimo-v2-6-pro","sourceUrl":"https://artificialanalysis.ai/models/mimo-v2-6-pro","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:8da9eb2a1cc3e8627d06a715f9daaa482e0ed1a39f6be819c78a2cf1e72c327c","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["8da9eb2a1cc3e8627d06a715f9daaa482e0ed1a39f6be819c78a2cf1e72c327c"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:deepseek:88d0e920aac0ddac","name":"DeepSeek V4 Pro 0813","sourceUrl":"https://artificialanalysis.ai/models/releases/deepseek-v4-pro","recommended":"max","chartReference":{"cost":0.674,"capability":36,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":36,"cost":0.674,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 36、$0.674/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":36,"boundaryProbePoints":1,"lowestObservedScore":20.4,"relativeFloor":28.8,"headroomFloor":21.959999999999997,"effectiveFloor":28.8,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":36,"cost":0.674,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"max","score":36,"cost":0.674,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["max"],"boundarySensitive":false,"boundaryProbes":[{"floor":27.8,"winner":"max"},{"floor":28.8,"winner":"max"},{"floor":29.8,"winner":"max"}],"excluded":[{"id":"non-reasoning","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["max"],"scenarioCount":11,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"max","label":"Max","capability":36,"cost":0.674,"evidence":{"sourceId":"83173329-c09a-41f1-a028-a282a5f908d7","originalName":"DeepSeek V4 Pro 0813 (Max)","sourceSlug":"deepseek-v4-pro","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":463.14,"method":"aa_published_estimate","configurationId":"83173329-c09a-41f1-a028-a282a5f908d7","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"deepseek-v4-pro","sourceUrl":"https://artificialanalysis.ai/models/deepseek-v4-pro","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:4654a4234dfb83f60fd7acb8452c6c0802861c626ad746cd3f0d42cde8038b0f","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4654a4234dfb83f60fd7acb8452c6c0802861c626ad746cd3f0d42cde8038b0f"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"non-reasoning","label":"Non-reasoning","capability":20.4,"cost":0.3555,"evidence":{"sourceId":"d4f5b250-be37-4801-85e6-32fda4179337","originalName":"DeepSeek V4 Pro 0813 (Non-reasoning)","sourceSlug":"deepseek-v4-pro-non-reasoning","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"non-reasoning","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":312.04,"method":"aa_published_estimate","configurationId":"d4f5b250-be37-4801-85e6-32fda4179337","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"deepseek-v4-pro-non-reasoning","sourceUrl":"https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:2c4937466d813523e7fbdf04aa6cf89f0076880240766dc474a58e09eb27562f","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["2c4937466d813523e7fbdf04aa6cf89f0076880240766dc474a58e09eb27562f"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:deepseek:1c2f02f74d295fa3","name":"DeepSeek V4 Flash 0731","sourceUrl":"https://artificialanalysis.ai/models/releases/deepseek-v4-flash","recommended":"max","chartReference":{"cost":0.2196,"capability":34.3,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":34.3,"cost":0.2196,"benchmark":"AA Intelligence Index"},"recommendationReason":"只有 Max 一组评测配置，推荐档就是这一档；没有其他档位可比。","recommendationPolicy":{"basis":"single-configuration","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"max","label":"Max","capability":34.3,"cost":0.2196,"evidence":{"sourceId":"fe4c0848-e284-4e52-a79d-cdc28392f1a9","originalName":"DeepSeek V4 Flash 0731 (Max)","sourceSlug":"deepseek-v4-flash","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":270.09,"method":"aa_published_estimate","configurationId":"fe4c0848-e284-4e52-a79d-cdc28392f1a9","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"deepseek-v4-flash","sourceUrl":"https://artificialanalysis.ai/models/deepseek-v4-flash","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:2cba7ade5d453604f302decd37740c2157985d0b2812c775962fe18fa0c08113","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["2cba7ade5d453604f302decd37740c2157985d0b2812c775962fe18fa0c08113"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:anthropic:ccb37fec2677f068","name":"Claude Haiku 5.5","sourceUrl":"https://artificialanalysis.ai/models/releases/claude-haiku-5-5","recommended":"xhigh","chartReference":{"cost":0.1238,"capability":41.2,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"xhigh","costEffort":"xhigh","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":43.4,"cost":0.2128,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：XHigh，得分 41.2、$0.1238/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":43.4,"boundaryProbePoints":1,"lowestObservedScore":29.5,"relativeFloor":34.72,"headroomFloor":30.89,"effectiveFloor":34.72,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"xhigh","score":41.2,"cost":0.1238,"timePenalty":[0,0],"worstRegret":0.7602131069680098,"robustWinner":false},"rankings":[{"id":"xhigh","score":41.2,"cost":0.1238,"timePenalty":[0,0],"worstRegret":0.7602131069680098,"robustWinner":false},{"id":"high","score":37.8,"cost":0.0794,"timePenalty":[0,0],"worstRegret":2.433860759493669,"robustWinner":false},{"id":"max","score":43.4,"cost":0.2128,"timePenalty":[0,0],"worstRegret":3.6337926002624883,"robustWinner":false}],"robustChoices":[],"boundarySensitive":false,"boundaryProbes":[{"floor":33.72,"winner":"xhigh"},{"floor":34.72,"winner":"xhigh"},{"floor":35.72,"winner":"xhigh"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"},{"id":"medium","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["high","xhigh"],"scenarioCount":23,"consistentAcrossScenarios":false,"robustWithinAllRegions":false,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":29.5,"cost":0.0245,"evidence":{"sourceId":"42908ae3-d3bd-4934-be19-fbff93d3fde8","originalName":"Claude Haiku 5.5 (Low)","sourceSlug":"claude-haiku-5-5-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"adaptive","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":66.92,"method":"aa_published_estimate","configurationId":"42908ae3-d3bd-4934-be19-fbff93d3fde8","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-haiku-5-5-low","sourceUrl":"https://artificialanalysis.ai/models/claude-haiku-5-5-low","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:65fe4777f8878d7ada1488449765f61216634d46a6d7a46d959dbece0bae5981","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["65fe4777f8878d7ada1488449765f61216634d46a6d7a46d959dbece0bae5981"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":34.5,"cost":0.047,"evidence":{"sourceId":"af26fff4-8120-41be-a5f0-86aaea023f20","originalName":"Claude Haiku 5.5 (Medium)","sourceSlug":"claude-haiku-5-5-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"adaptive","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":128.92,"method":"aa_published_estimate","configurationId":"af26fff4-8120-41be-a5f0-86aaea023f20","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-haiku-5-5-medium","sourceUrl":"https://artificialanalysis.ai/models/claude-haiku-5-5-medium","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:32c3970ae3d2452c9885855e2257fc2f82027e4734868bccf76cd294675f0593","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["32c3970ae3d2452c9885855e2257fc2f82027e4734868bccf76cd294675f0593"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":37.8,"cost":0.0794,"evidence":{"sourceId":"20829974-af06-497e-9f42-77fb958d43b9","originalName":"Claude Haiku 5.5 (High)","sourceSlug":"claude-haiku-5-5-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"adaptive","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":209.94,"method":"aa_published_estimate","configurationId":"20829974-af06-497e-9f42-77fb958d43b9","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-haiku-5-5-high","sourceUrl":"https://artificialanalysis.ai/models/claude-haiku-5-5-high","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:d2b5cd8ac4dc84e1febd7deaa87bb1f1d6b245a9ede0b7c136add3886d839b50","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["d2b5cd8ac4dc84e1febd7deaa87bb1f1d6b245a9ede0b7c136add3886d839b50"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":41.2,"cost":0.1238,"evidence":{"sourceId":"41b5bf5e-daf6-42d9-b8d1-60e340a3028d","originalName":"Claude Haiku 5.5 (Xhigh)","sourceSlug":"claude-haiku-5-5-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"adaptive","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":295.36,"method":"aa_published_estimate","configurationId":"41b5bf5e-daf6-42d9-b8d1-60e340a3028d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-haiku-5-5-xhigh","sourceUrl":"https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:c779d11a478ebd7339f0ac6458aa5b09a99af355fd4e52ad8224a2f74d557453","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["c779d11a478ebd7339f0ac6458aa5b09a99af355fd4e52ad8224a2f74d557453"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":43.4,"cost":0.2128,"evidence":{"sourceId":"7d064a11-6e78-4030-b535-cdf940615041","originalName":"Claude Haiku 5.5 (Max)","sourceSlug":"claude-haiku-5-5","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"adaptive","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":424.77,"method":"aa_published_estimate","configurationId":"7d064a11-6e78-4030-b535-cdf940615041","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-haiku-5-5","sourceUrl":"https://artificialanalysis.ai/models/claude-haiku-5-5","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:577f4364243f3ae1c4f55b4e92eb1cdebb6c36b4d4eb2b1c9dae30a6d8df9647","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["577f4364243f3ae1c4f55b4e92eb1cdebb6c36b4d4eb2b1c9dae30a6d8df9647"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"sonnet-5.5","name":"Claude Sonnet 5.5","sourceUrl":"https://artificialanalysis.ai/models/releases/claude-sonnet-5-5","recommended":"high","chartReference":{"cost":0.8844,"capability":46.8,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"high","costEffort":"high","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":56,"cost":5.4609,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：High，得分 46.8、$0.8844/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":56,"boundaryProbePoints":1,"lowestObservedScore":35.9,"relativeFloor":44.800000000000004,"headroomFloor":37.91,"effectiveFloor":44.800000000000004,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"high","score":46.8,"cost":0.8844,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"high","score":46.8,"cost":0.8844,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},{"id":"xhigh","score":51.9,"cost":2.012,"timePenalty":[0,0],"worstRegret":52.252068965517225,"robustWinner":false},{"id":"max","score":56,"cost":5.4609,"timePenalty":[0,0],"worstRegret":223.5702586206896,"robustWinner":false}],"robustChoices":["high"],"boundarySensitive":false,"boundaryProbes":[{"floor":43.800000000000004,"winner":"high"},{"floor":44.800000000000004,"winner":"high"},{"floor":45.800000000000004,"winner":"high"}],"excluded":[{"id":"low","reason":"below-declared-relative-score-protection"},{"id":"medium","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["high"],"scenarioCount":23,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"low","label":"Low","capability":35.9,"cost":0.3452,"evidence":{"sourceId":"bbc2ffea-cf6b-43be-8c41-1769347e234d","originalName":"Claude Sonnet 5.5 (Low, Default Fallback)","sourceSlug":"claude-sonnet-5-5-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":95.74,"method":"aa_published_estimate","configurationId":"bbc2ffea-cf6b-43be-8c41-1769347e234d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-sonnet-5-5-low","sourceUrl":"https://artificialanalysis.ai/models/claude-sonnet-5-5-low","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:f23928edc35d41b5e96abfc75d6283d67b4786b687997863f1cbf5f57a97c32b","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["f23928edc35d41b5e96abfc75d6283d67b4786b687997863f1cbf5f57a97c32b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":40.8,"cost":0.4833,"evidence":{"sourceId":"042e8a37-fcd0-40f4-8f41-99215fb68eda","originalName":"Claude Sonnet 5.5 (Medium, Default Fallback)","sourceSlug":"claude-sonnet-5-5-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":130.6,"method":"aa_published_estimate","configurationId":"042e8a37-fcd0-40f4-8f41-99215fb68eda","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-sonnet-5-5-medium","sourceUrl":"https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","sourceObservedAt":"2026-10-15T13:31:58.220Z","sourceEvidenceId":"aa-public-model-page:162f87cb9b831b518ded1fe860ce236f0736d0ea78c1c04a904738c6ca5ed15d","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["162f87cb9b831b518ded1fe860ce236f0736d0ea78c1c04a904738c6ca5ed15d"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":46.8,"cost":0.8844,"evidence":{"sourceId":"268525e2-f873-4178-bc07-e6a0ee1f002d","originalName":"Claude Sonnet 5.5 (High, Default Fallback)","sourceSlug":"claude-sonnet-5-5-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":238.11,"method":"aa_published_estimate","configurationId":"268525e2-f873-4178-bc07-e6a0ee1f002d","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-sonnet-5-5-high","sourceUrl":"https://artificialanalysis.ai/models/claude-sonnet-5-5-high","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:483ad4ec152187e4b9f454b24f46d7880ea3748da65fffc3e73e40b33f1b1e2c","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["483ad4ec152187e4b9f454b24f46d7880ea3748da65fffc3e73e40b33f1b1e2c"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":51.9,"cost":2.012,"evidence":{"sourceId":"975aeaef-50bd-4a65-9c73-f1015adcd4d4","originalName":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","sourceSlug":"claude-sonnet-5-5-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":472.78,"method":"aa_published_estimate","configurationId":"975aeaef-50bd-4a65-9c73-f1015adcd4d4","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-sonnet-5-5-xhigh","sourceUrl":"https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:290bb78444856f231451dfca6208aad5ac6f9df6a2f162f44ac08a248571b979","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["290bb78444856f231451dfca6208aad5ac6f9df6a2f162f44ac08a248571b979"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":56,"cost":5.4609,"evidence":{"sourceId":"b171d979-5ec1-45de-b8b9-db1ee65e77ec","originalName":"Claude Sonnet 5.5 (Max, Default Fallback)","sourceSlug":"claude-sonnet-5-5","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"adaptive","fallbackPolicy":"default"},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":938.48,"method":"aa_published_estimate","configurationId":"b171d979-5ec1-45de-b8b9-db1ee65e77ec","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"claude-sonnet-5-5","sourceUrl":"https://artificialanalysis.ai/models/claude-sonnet-5-5","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:4bb3055964a10330b320bcf24fb49848069f87fedebc47e10953c58585171930","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["4bb3055964a10330b320bcf24fb49848069f87fedebc47e10953c58585171930"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:minimax:9250ce592a868329","name":"MiniMax-M3","sourceUrl":"https://artificialanalysis.ai/models/releases/minimax-m3","recommended":"observed","chartReference":{"cost":0.5076,"capability":29.2,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":29.2,"cost":0.5076,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":29.2,"cost":0.5076,"evidence":{"sourceId":"277f939a-985b-4b37-859d-b3eabc7c0b26","originalName":"MiniMax-M3","sourceSlug":"minimax-m3","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":502.13,"method":"aa_published_estimate","configurationId":"277f939a-985b-4b37-859d-b3eabc7c0b26","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"minimax-m3","sourceUrl":"https://artificialanalysis.ai/models/minimax-m3","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:29f218409531c1305afa06054fbffa4f86a9926668ff3a49917be3a1de99c88d","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["29f218409531c1305afa06054fbffa4f86a9926668ff3a49917be3a1de99c88d"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:deepseek:b75e488c1a8d0e9d","name":"DeepSeek V4.1 Flash","sourceUrl":"https://artificialanalysis.ai/models/releases/deepseek-v4-1-flash-non-reasoning","recommended":"max","chartReference":{"cost":0.2652,"capability":39.5,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":39.5,"cost":0.2652,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 39.5、$0.2652/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"comparison-only","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":39.5,"boundaryProbePoints":1,"lowestObservedScore":24.7,"relativeFloor":31.6,"headroomFloor":26.18,"effectiveFloor":31.6,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":39.5,"cost":0.2652,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true},"rankings":[{"id":"max","score":39.5,"cost":0.2652,"timePenalty":[0,0],"worstRegret":0,"robustWinner":true}],"robustChoices":["max"],"boundarySensitive":false,"boundaryProbes":[{"floor":30.6,"winner":"max"},{"floor":31.6,"winner":"max"},{"floor":32.6,"winner":"max"}],"excluded":[{"id":"non-reasoning","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["max"],"scenarioCount":11,"consistentAcrossScenarios":true,"robustWithinAllRegions":true,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"non-reasoning","label":"Non-reasoning","capability":24.7,"cost":0.1467,"evidence":{"sourceId":"73c96f99-0de3-4668-a4bd-b106a0fe18f2","originalName":"DeepSeek V4.1 Flash (Non-reasoning)","sourceSlug":"deepseek-v4-1-flash-non-reasoning","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"non-reasoning","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":206.43,"method":"aa_published_estimate","configurationId":"73c96f99-0de3-4668-a4bd-b106a0fe18f2","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"deepseek-v4-1-flash-non-reasoning","sourceUrl":"https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","sourceObservedAt":"2026-10-15T07:16:30.370Z","sourceEvidenceId":"aa-public-model-page:2c09db2d3aba797105addee3797af66efc4168545d112a57e0d2439350550698","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["2c09db2d3aba797105addee3797af66efc4168545d112a57e0d2439350550698"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":39.5,"cost":0.2652,"evidence":{"sourceId":"dbe7c625-3100-4463-b479-a228c41f75dd","originalName":"DeepSeek V4.1 Flash (Max)","sourceSlug":"deepseek-v4-1-flash","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":350.1,"method":"aa_published_estimate","configurationId":"dbe7c625-3100-4463-b479-a228c41f75dd","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"deepseek-v4-1-flash","sourceUrl":"https://artificialanalysis.ai/models/deepseek-v4-1-flash","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:1681e923b7a99cc5cbb33a5aff48bef0879bec9f8fdf6fd67ec0e93695447eda","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["1681e923b7a99cc5cbb33a5aff48bef0879bec9f8fdf6fd67ec0e93695447eda"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"auto:tencent:f1330786ec5f15d5","name":"Hy3","sourceUrl":"https://artificialanalysis.ai/models/releases/hy3","recommended":"observed","chartReference":{"cost":0.0718,"capability":25.3,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":25.3,"cost":0.0718,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":25.3,"cost":0.0718,"evidence":{"sourceId":"b23e6c69-96e5-44c9-8f58-4b42e0c399d5","originalName":"Hy3","sourceSlug":"hy3","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":524.24,"method":"aa_published_estimate","configurationId":"b23e6c69-96e5-44c9-8f58-4b42e0c399d5","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"hy3","sourceUrl":"https://artificialanalysis.ai/models/hy3","sourceObservedAt":"2026-10-15T10:11:20.793Z","sourceEvidenceId":"aa-public-model-page:c16974df3af8158fd26447f229194ffbae635aa71522a3d060dc5b2312dd4b10","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["c16974df3af8158fd26447f229194ffbae635aa71522a3d060dc5b2312dd4b10"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]},{"id":"luna","name":"GPT-6 Luna","sourceUrl":"https://artificialanalysis.ai/models/releases/gpt-6-luna","recommended":"max","chartReference":{"cost":0.0678,"capability":38.1,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"max","costEffort":"max","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"max","capability":38.1,"cost":0.0678,"benchmark":"AA Intelligence Index"},"recommendationReason":"均衡建议 v4：Max，得分 38.1、$0.0678/任务。绝对费用与倍率共同参与；质量门槛兼顾最高得分的80%与至少10%同模型可调增量。AA 生成耗时及独立 Agent 耗时都不参与默认推荐。这是暂定产品规则，不是官方推荐或任务成功保证。","recommendationPolicy":{"releaseVersion":"0.21.115","version":"balanced-coding-absolute-v4","activeSettings":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","status":"experimental-scenario-policy-not-independent-validation","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"boundaryProbePoints":1,"parameterProvenance":{"priceWeights":"Joint nonnegative absolute-cost and log-cost region constrained by the owner's recorded Sol comparisons; no family overrides or hand-picked midpoint.","retainedScoreFraction":"Existing provisional benchmark-score floor at 80% of the same-model maximum, not retained real-world task success.","minimumHeadroomFraction":"Existing common protection: preserve at least 10% of observed within-model score improvement across priced native efforts.","timePolicy":"No time field enters default candidate admission, regret, Pareto, tie-break or explanation. AA generation estimate is a display metric; TB/AA Agent wall time remains independent detail evidence."},"sensitivity":{"headroomFractions":[0,0.05,0.1,0.15,0.2,0.25,0.3],"retainedScoreFractions":[0.75,0.8,0.82,0.84,0.85]},"automaticRoutingEligible":false},"calibrationPreferences":{"version":"owner-preferences-2026-09-27","sourceSnapshotId":"aa-1b7eb89c-2abd-445e-aece-3caafcfb78cd","scope":"AA API Intelligence Index v4.3 / USD per benchmark task; patch identifier not supplied by API","status":"preference-samples-not-independent-validation","timePolicyStatus":"inactive_legacy_proposal_not_used_by_balanced_v4","timePolicy":"INACTIVE historical proposal from the earlier bounded-time scenario. Balanced v4 passes no timePolicy context and uses only AA score and monetary task cost; this text is not a calibrated preference or active ranking rule.","preferences":[{"preferred":{"id":"sol:high","score":42.8,"cost":0.3746},"alternative":{"id":"sol:max","score":47.5,"cost":1.0564},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:high","score":42.8,"cost":0.3746},"provenance":"User explicitly confirmed on 2026-09-27"},{"preferred":{"id":"sol:medium","score":39.8,"cost":0.2482},"alternative":{"id":"sol:low","score":33.9,"cost":0.1322},"provenance":"User explicitly confirmed on 2026-09-27"}],"versionResolutionNote":"The referenced methodology webpage used v4.3.2, but the frozen API identifies v4.3 only. Numeric preference samples are unchanged; patch compatibility is not asserted."},"referenceEffort":{"policy":{"id":"bughunt-effort-reference-v1","version":"2026-10-13.1","scope":"debugging-benchmark-only","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"selection":"least-source-cost-after-quality-protection","defaultRecommendationChanged":false,"automaticRoutingEligible":false},"formula":"floor = max(0.80 * max(score), min(score) + 0.10 * (max(score) - min(score))); choose least complete-run source cost at or above floor","inputs":"Reviewed exact-model Bug Hunt cohorts only; bugs fixed with a meaningful zero, one harness/route and cost kind. Source list-price estimates and billed costs are never pooled.","updates":"Numeric corrections refresh from normal source snapshots; protocol, context, membership or repetition changes require cohort review. Stale/missing sources withhold the reference.","limits":"Provisional debugging-workload contrast. With two unequal-score rungs the headroom guard forces the higher rung. Unknown CLI/context and single-run caveats remain explicit; not a task-success guarantee or an executable native setting.","compatibility":"Optional referenceEffort on Index models and /v1/index-metrics profiles; read via MCP get_model_metrics. The size-bounded client digest omits full references. AA-native balanced recommendation, efforts, ability/cost/time coordinates and default routing are unchanged."},"activeDefaultTimePolicy":"none; historical calibrationPreferences.timePolicy is inactive","status":"versioned-policy","transparency":{"releaseVersion":"0.21.115","displayMetrics":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"crossModelTime":{"version":"protocol-calibrated-time-v1","minimumModels":3,"minimumConfigurations":6,"highErrorFactor":2},"timeFieldUsage":{"crossModel":"indexMetrics.taskTimeCrossModel.value","withinModel":"indexMetrics.taskTimeWithinModel","sourceReported":"indexMetrics.taskTimeMultiplier","seconds":"indexMetrics.taskTimeCrossModel.sourceSeconds","legacySeconds":"indexMetrics.taskSeconds","estimates":"indexMetrics.taskTimeCrossModel.estimateValue","rule":"Use one global calibration before filtering. Cross-model values are experimental dimensionless coordinates; seconds retain their own task boundary. Within-model ratios require the same protocol/component. Never place models using taskTimeWithinModel."},"speed":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"speedFieldUsage":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"versions":{"abilityIndex":"unified-v03-estimator-0.1.0-internal","abilityDisplay":"0.3","abilityWithinModel":"same-source-chain-v1","taskTimeWithinModel":"same-protocol-time-v1","taskTimeCrossModel":"protocol-calibrated-time-v1","taskTimeEstimate":"task-time-evidence-v2","recommendation":"balanced-coding-absolute-v4"},"repository":"https://github.com/guanbear/modelplans.ai","repositoryVisibilityAtDisclosure":"private","disclosureDate":"2026-09-30","documentation":"https://github.com/guanbear/modelplans.ai/blob/feat/index-v03-publication/index/unified_v03/README.md","active":{"version":"balanced-coding-absolute-v4","qualityInput":"observed AA score for the selected native configuration","costInput":"same-configuration mean USD per AA benchmark task","formula":"U = Q - a*C - b*log2(C/Cref), Cref=1 USD/AA task","retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"guard":"max(0.8*Qmax, Qmin + 0.1*(Qmax-Qmin)) on observed priced AA scores only","priceCalibration":"three recorded owner Sol preferences; nonnegative feasible price region; minimize worst regret","officialDefaults":"reference only","timeInput":"none"},"experimentalEffort":{"version":"partial-effort-gains-v0.2-experimental","baselineId":"effort-v02-first-2026-09-30","scope":"offline candidate; not the currently deployed default","sourceBudgets":{"AA":0.5,"LiveBench Coding":0.125,"LiveBench Agentic":0.125,"CursorBench":0.25},"missing":"no source penalty or budget redistribution; singleton Max cannot infer effort gains; unknown calibrated anchors/prices abstain","assumptions":"source weights/unit bridge and AA score guards are declared product assumptions; real-task optimality is unverified"},"publicCodePreparation":{"licenseScope":"MIT for allowlisted original algorithm package, derived parameters and synthetic example; upstream datasets retain their own terms","exportCommand":"python3 index/export_public.py --output /tmp/modelplan-index-v02","validation":"frozen baseline/code/preferences/scales; offline replay, sparse-data invariants and separate sensitivity scenarios"}},"websiteDisplay":{"releaseVersion":"0.21.115","version":"0.3","endpoint":"/api/visualization/index-models","api":"/v1/index-metrics","method":{"version":"0.3","methodId":"unified-v03-estimator-0.1.0-internal","status":"experimental","sourceBudgets":{"aa":{"label":"AA综合","publisher":"AA","ability":0.3,"cost":0.2,"time":0},"aa-agent":{"label":"AA Coding Agent","publisher":"AA","ability":0.2,"cost":0.3,"time":0.6},"lb-coding":{"label":"LiveBench Coding","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-agentic":{"label":"LiveBench Agentic","publisher":"LiveBench","ability":0.1,"cost":0,"time":0},"lb-cost":{"label":"LiveBench任务费用","publisher":"LiveBench","ability":0,"cost":0.1,"time":0},"cursor":{"label":"CursorBench","publisher":"Cursor","ability":0.1,"cost":0.15,"time":0},"tb":{"label":"Terminal-Bench","publisher":"TB","ability":0.1,"cost":0.15,"time":0.25},"swebench":{"label":"SWE-bench","publisher":"SWE","ability":0.04,"cost":0.03,"time":0},"aider":{"label":"Aider","publisher":"Aider","ability":0.02,"cost":0.02,"time":0},"livecodebench":{"label":"LiveCodeBench","publisher":"LCB","ability":0.02,"cost":0,"time":0},"bughunt":{"label":"Bug Hunt","publisher":"Bug Hunt","ability":0.02,"cost":0.05,"time":0.15}},"references":{"aa":{"ability":{"status":"ready","mu":45.044444444444444,"sd":4.781703813289223,"geometricReference":null},"cost":{"status":"ready","mu":0.4491124177869664,"sd":1.4180701137552665,"geometricReference":1.5669207970669967},"seconds":{"status":"insufficient-reference"}},"aa-agent":{"ability":{"status":"ready","mu":58.65231477076087,"sd":9.185549683865638,"geometricReference":null},"cost":{"status":"ready","mu":1.4959394727153073,"sd":1.508417570970172,"geometricReference":4.463527946409263},"seconds":{"status":"ready","mu":7.70346257527284,"sd":0.535160611619162,"geometricReference":2216.0078167447414}},"lb-coding":{"ability":{"status":"ready","mu":84.15294642857143,"sd":4.131638835182962,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-agentic":{"ability":{"status":"ready","mu":59.4504126984127,"sd":7.331316931406503,"geometricReference":null},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"lb-cost":{"ability":{"status":"insufficient-reference"},"cost":{"status":"ready","mu":-1.0110654847069085,"sd":1.0254070298772795,"geometricReference":0.36383111648661043},"seconds":{"status":"insufficient-reference"}},"cursor":{"ability":{"status":"ready","mu":44.56666666666666,"sd":7.0521076439758215,"geometricReference":null},"cost":{"status":"ready","mu":1.7998967091343137,"sd":0.7612549616257507,"geometricReference":6.049022623359992},"seconds":{"status":"insufficient-reference"}},"tb":{"ability":{"status":"ready","mu":36.843333333333334,"sd":16.45829392400345,"geometricReference":null},"cost":{"status":"ready","mu":2.2593856984828053,"sd":0.8552892284718037,"geometricReference":9.577204068347337},"seconds":{"status":"ready","mu":8.386697437198304,"sd":0.35953490975737623,"geometricReference":4388.301096474238}},"swebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"aider":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"livecodebench":{"ability":{"status":"insufficient-reference"},"cost":{"status":"insufficient-reference"},"seconds":{"status":"insufficient-reference"}},"bughunt":{"ability":{"status":"ready","mu":29.13227513227513,"sd":12.04373440054195,"geometricReference":null},"cost":{"status":"ready","mu":3.356130136975273,"sd":1.0034045668975784,"geometricReference":28.677995947294892},"seconds":{"status":"ready","mu":8.445881749405338,"sd":0.7079230021833092,"geometricReference":4655.859204661198}}},"abilityFormula":"S = 50 + 10 * sum(w*r*z) / sum(w*r); z = (x-mu)/sigma","expenseFormula":"E = exp(sum(w*r*log(C/Cref)) / sum(w*r))","taskTimeFormula":"Tindex = T / Tref_source; AA Agent > Terminal-Bench > eligible Bug Hunt; boundaries never averaged","completenessFormula":"D = 100 * sum(admitted direct source budgets) / sum(all declared budgets)","recommendationPolicy":"balanced-coding-absolute-v4; unchanged; time excluded","anchorsEnabled":false},"quality":"frozen-scale multisource estimate; not a probability or proportional ability","expense":"source workload reference multiplier, not USD or subscription quota","taskTime":"explicit per-model protocol coordinate with overlap-calibrated source scale and error flags; source seconds retain their own task boundary","outputThroughput":{"policy":{"version":"channel-speed-evidence-v1","maximumChangesPerConfiguration":32,"measurementField":"median_output_tokens_per_second","diagnosticChannel":"api","quantile":"empirical midrank: (count strictly below + 0.5 * count equal) / population size","population":"native recommended configurations with both observed cross-model task coordinates and measured API throughput; before request filters","clientRule":"taskTimeQuantile <= 0.25 and throughputQuantile >= 2/3; clients decide eligibility using provenance and uncertainty flags"},"fields":{"throughput":"indexMetrics.outputThroughput.value","channel":"indexMetrics.outputThroughput.channel","taskTimeQuantile":"indexMetrics.speedDiagnostics.taskTimeQuantile","throughputQuantile":"indexMetrics.speedDiagnostics.throughputQuantile","throughputVsMedian":"indexMetrics.speedDiagnostics.throughputVsMedian","claims":"model.throughputClaims"},"source":"AA per-configuration Free API performance; subscription announcements/reports are separate reviewed claims"},"recommendation":"recommendation uses native-AA quality/USD calibration, time excluded"},"chartReference":{"scope":"legacy /api/visualization/models contract; current website uses websiteDisplay","capability":"selected-effort AA Intelligence score","cost":"same selected-effort AA Intelligence benchmark task cost","capsule":"selected effort, initially recommended","representsSingleConfiguration":true,"highCapabilityReference":"one separately labeled, highest-scoring priced native effort of the active model","timing":"default Z is AA's per-configuration Intelligence Index decode-time estimate; legacy TB trial and Coding Agent wall time are detail-only references","generationTime":{"metric":"aa_index_generation_estimate","methodologyVersion":"aa-index-generation-v4","source":"Dated AA public Index chart labels; exact first-party model pages; a separate, labeled ModelPlans same-model median-rate estimate only after leave-one-effort-out validation","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=claude-opus-5-5-medium%2Cgpt-6-astra-medium%2Cclaude-fable-5-1-medium%2Cgpt-6-sol-medium%2Cgpt-6-luna-xhigh%2Cclaude-sonnet-5-high%2Cgpt-6-luna-non-reasoning%2Cgpt-6-luna-low%2Cgpt-6-luna-medium%2Cgpt-6-luna-high%2Cgpt-6-luna%2Cclaude-sonnet-5-non-reasoning%2Cclaude-sonnet-5-low%2Cclaude-sonnet-5-medium%2Cclaude-sonnet-5-xhigh%2Cclaude-sonnet-5%2Cgpt-6-sol-non-reasoning%2Cgpt-6-sol-low%2Cgpt-6-sol-high%2Cgpt-6-sol-xhigh%2Cgpt-6-sol%2Cclaude-opus-5-5-low%2Cclaude-opus-5-5-high%2Cclaude-opus-5-5-xhigh%2Cclaude-opus-5-5%2Cgpt-6-astra-low%2Cgpt-6-astra-high%2Cgpt-6-astra-xhigh%2Cgpt-6-astra%2Cclaude-fable-5-1-low%2Cclaude-fable-5-1-high%2Cclaude-fable-5-1-xhigh%2Cclaude-fable-5-1%2Cglm-5-3-low%2Cglm-5-3#speed","chartVersion":"v4.3.2","sourceObservedAt":"2026-09-28T06:17:38Z","modelPageAuditObservedAt":"2026-09-28T09:03:15.124Z","modelPageAuditSource":"Public rendered model pages, exact model ID and Index major.minor checked; current missing values remain null","modelPageCandidateMaxAgeDays":7,"displayedPrecisionMinutes":0.1,"coveredConfigurations":32,"missing":"null; no TB/Agent wall or 500-token response fallback. Sol/Luna Medium have exact weighted tokens but no current matching default speed; Sonnet non-reasoning has speed but no Index tokens/task.","excludedComponents":["TTFT","tools","harness setup","verification"],"sampling":"Public chart values may refresh independently of the Free AA score/cost snapshot; patch correspondence is not guaranteed.","metricSeparation":"Time per Intelligence Index Task is a weighted decode-time estimate; End-to-End Response Time is a separate 500-answer-token response metric.","sameModelEstimate":{"policy":{"version":"same-model-output-time-v2","maxSourceAgeMs":604800000,"maxObservationSpanMs":900000,"minTimedAnchors":4,"minBracketedHoldouts":2,"maxMeanAbsolutePercentError":0.05,"maxSingleAbsolutePercentError":0.1,"maxNeighborRateSpread":0.15,"maxComparableHistoricalError":0.2},"formula":"T_hat(e) = N_e * median_i(T_i / N_i), only same model release, reasoning mode and AA Index patch; target token count must lie between timed anchors","status":"AA target direct time remains missing. Admitted ModelPlans estimates use a dashed Z trail; a historical lead with unknown patch or performance window leaves a visible trialEstimate with null quantitative Z. Numeric difference is not called prediction error until comparability is confirmed.","source":"data/aa-same-model-output-time.json and versioned admission decision"},"visualGhost":"UI-only soft shadow when the selected configuration lacks admitted time. It has no metric value, zero-time meaning, speed rank, API generationTime.valueSeconds or role in the balanced recommendation."},"legacyTaskTime":"Deprecated `time` and `taskTime` retain their original TB/Agent task-time meaning; never relabel them as generation time.","missingSelectedMetric":"not plotted; never substituted with Max or another effort"},"sourceRoles":"Field ownership and admission rules are published by /v1/benchmark-sources.","capability":{"method":"Keep the source score and unit within its dataset/version/subset/harness cohort. No cross-benchmark arithmetic average.","missing":"null, never zero","copiedEvidence":"One originating observation; references add no votes."},"recommendation":{"method":"For verified general/balanced AA curves, enforce same-model score protection and explicit cost constraints, then minimize regret with absolute and relative price penalties. No time field enters default admission, regret, Pareto, tie-break or explanation. Specialist task-cohort policies are separate.","balancedScoreRetention":0.8,"balancedHeadroomRetention":0.1,"scope":"Default general/balanced recommendations use verified model effort curves independently; specialist and explicit legacy policies retain their own protocols.","formula":"U = Q - a*C - b*log2(C/Cref); default time preference is absent","selection":"Minimize maximum regret over the joint feasible nonnegative price-weight region; AA generation time and independent Agent time are not ranking inputs. Automatic routing remains disabled.","balancedExplanation":"Default general/balanced uses an 80% same-model score floor plus at least 10% of observed within-model effort score gain, then joint absolute and relative price regret. Both floors are provisional benchmark-score policies, not task-success evidence. Time is absent from default ranking; task-specific evidence remains separately labeled.","defaultTolerance":{"general":3,"coding":3,"agentic":3,"debugging":5,"terminal":3,"software-repair":3,"code-editing":3,"code-generation":3},"units":{"general":"AA index points","coding":"AA coding index points","agentic":"AA agentic index points","debugging":"bugs fixed / 105","terminal":"accuracy percentage points","software-repair":"resolved percentage points","code-editing":"pass_rate_2 percentage points","code-generation":"pass@1 percentage points"},"qualityPolicy":"Highest raw score; small differences are not statistical significance.","missingTime":"The default balanced policy does not use time. An explicit specialist speed policy requires measured, comparable task duration and abstains when it is missing.","crossTaskCost":"Reference only; excluded from budget constraints and cost optimization."},"uncertainty":{"confidenceInterval":"Preserve published intervals; never infer repeated-run confidence from trial count.","unknownHarnessVersion":"Provisional evidence; shown explicitly.","automation":"Source ingestion is automatic after schema and anomaly checks. Recommendation consumers must inspect freshness, evidence status, constraints and policy version."},"changeControl":{"method":"Versioned code and regression fixtures; changes require review.","identity":"Explicit source relationships or reviewed mappings.","entitlements":"Candidates plus evidence-bound approval before formal publication."},"validation":["missing-is-not-zero","cohorts-never-pooled","unknown-effort-not-fabricated","measured-time-only","trial-count-not-repeats","last-good-on-failure"],"policyHash":"63eee71a5c65ec74aab3c3d2bb0e00a1903f4916eb5da07b571e27f4eecf93d2","experimentalPreview":{"version":"balanced-coding-absolute-v4","endpoint":"/v1/recommendation-preview","formula":"U = Q - a*C - b*log2(C/Cref) in the active profile; no time term","status":"Default endpoint explains the active policy; explicit legacy parameters request a non-routing scenario comparison only.","timeEstimates":"Ordinal-effort interpolation is research-only. It never enters the active recommendation or automatic routing.","automaticRoutingEligible":false},"decision":{"policy":{"id":"balanced-coding-absolute-v4","version":"2026-09-28.2","fittedPriceParameterCount":2,"modelExceptions":false,"status":"experimental-scenario-policy"},"status":"boundary-sensitive","priceRegion":{"status":"bounded","constraints":[{"a":-1,"b":0,"limit":0},{"a":0,"b":-1,"limit":0},{"a":-0.6818,"b":-1.4957334019240476,"limit":-4.700000000000003},{"a":-0.12639999999999998,"b":-0.5938477841126888,"limit":-3},{"a":0.11599999999999999,"b":0.9087809386226648,"limit":5.899999999999999}],"vertices":[{"absolute":0,"relative":5.051799602961419,"epsilon":0},{"absolute":0,"relative":6.492213633950061,"epsilon":0},{"absolute":23.734177215189877,"relative":0,"epsilon":0},{"absolute":50.86206896551723,"relative":0,"epsilon":0}]},"qualityProtection":{"retainedScoreFraction":0.8,"minimumHeadroomFraction":0.1,"referenceBestScore":38.1,"boundaryProbePoints":1,"lowestObservedScore":18.5,"relativeFloor":30.480000000000004,"headroomFloor":20.46,"effectiveFloor":30.480000000000004,"meaning":"benchmark score policy, not task-success evidence"},"context":{"protocol":{"score":"aa-intelligence-v4.3","cost":"aa-intelligence-usd-per-task-v4.3"}},"compromise":{"id":"max","score":38.1,"cost":0.0678,"timePenalty":[0,0],"worstRegret":2.754470317535798,"robustWinner":false},"rankings":[{"id":"max","score":38.1,"cost":0.0678,"timePenalty":[0,0],"worstRegret":2.754470317535798,"robustWinner":false},{"id":"xhigh","score":34.6,"cost":0.0422,"timePenalty":[0,0],"worstRegret":2.892405063291136,"robustWinner":false},{"id":"high","score":32.9,"cost":0.029,"timePenalty":[0,0],"worstRegret":4.279113924050634,"robustWinner":false}],"robustChoices":[],"boundarySensitive":true,"boundaryProbes":[{"floor":29.480000000000004,"winner":"xhigh"},{"floor":30.480000000000004,"winner":"max"},{"floor":31.480000000000004,"winner":"max"}],"excluded":[{"id":"none","reason":"below-declared-relative-score-protection"},{"id":"low","reason":"below-declared-relative-score-protection"},{"id":"medium","reason":"below-declared-relative-score-protection"}],"automaticRoutingEligible":false,"timeStatus":"not-configured-not-assumed-zero-seconds","scoreUncertainty":"point-values-present-not-confidence-bounds"}},"calibrationPreview":{"status":"preview","candidateEfforts":["xhigh","max"],"scenarioCount":27,"consistentAcrossScenarios":false,"robustWithinAllRegions":false,"automaticRoutingEligible":false,"reason":"当前星标只按 AA 综合能力和对应任务成本折中；生成时间及独立 Agent 时间不进入候选、regret 或 Pareto。最低保留10%同模型可调得分增量；扰动与质量边界只作敏感性检查。"},"configurations":[{"effort":"none","label":"None","capability":18.5,"cost":0.0113,"evidence":{"sourceId":"63cb48ea-0b0c-4e07-8ded-6d28dec4fa28","originalName":"GPT-6 Luna (Non-reasoning)","sourceSlug":"gpt-6-luna-non-reasoning","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"none","thinkingMode":"disabled","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":33.68,"method":"aa_published_estimate","configurationId":"63cb48ea-0b0c-4e07-8ded-6d28dec4fa28","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna-non-reasoning","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:2a0d33fc9949978f98dcf4464c38072f1e9e8b296c71af746651577e51eb3a8b","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["2a0d33fc9949978f98dcf4464c38072f1e9e8b296c71af746651577e51eb3a8b"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"low","label":"Low","capability":21.5,"cost":0.0045,"evidence":{"sourceId":"bf9708d8-933d-44f6-affa-b696a7a650c9","originalName":"GPT-6 Luna (Low)","sourceSlug":"gpt-6-luna-low","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"low","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":17.75,"method":"aa_published_estimate","configurationId":"bf9708d8-933d-44f6-affa-b696a7a650c9","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna-low","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna-low","sourceObservedAt":"2026-10-14T22:32:30.916Z","sourceEvidenceId":"aa-public-model-page:f92a7c029c56d90b5be483c33279df4b3393eb3bb33b08e418d1094b276522ab","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["f92a7c029c56d90b5be483c33279df4b3393eb3bb33b08e418d1094b276522ab"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"medium","label":"Med","capability":29.9,"cost":0.0175,"evidence":{"sourceId":"36667da0-9222-4967-adbb-f7efa15ad213","originalName":"GPT-6 Luna (Medium)","sourceSlug":"gpt-6-luna-medium","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"medium","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":null,"method":"unavailable","configurationId":"36667da0-9222-4967-adbb-f7efa15ad213","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna-medium","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna-medium","sourceObservedAt":"2026-10-15T09:31:40.993Z","sourceEvidenceId":"aa-public-model-page:54eb9fa32e212d88d965421faf8814123db4a56b70d125c2c784d479092a0e09","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["54eb9fa32e212d88d965421faf8814123db4a56b70d125c2c784d479092a0e09"],"unavailableReason":"no_public_chart_value","sameModelEstimateStatus":"target_identity_or_score_mismatch"},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"high","label":"High","capability":32.9,"cost":0.029,"evidence":{"sourceId":"7d1229bc-deea-4717-ba5b-f59a35d991e3","originalName":"GPT-6 Luna (High)","sourceSlug":"gpt-6-luna-high","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"high","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":162.17,"method":"aa_published_estimate","configurationId":"7d1229bc-deea-4717-ba5b-f59a35d991e3","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna-high","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna-high","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:93b20afc6b78a274fc7ed1249dc52379f2157dd27daf0bedd194a97d684c5851","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["93b20afc6b78a274fc7ed1249dc52379f2157dd27daf0bedd194a97d684c5851"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"xhigh","label":"XHigh","capability":34.6,"cost":0.0422,"evidence":{"sourceId":"19813eb2-460a-475c-af65-810bb8660fec","originalName":"GPT-6 Luna (Xhigh)","sourceSlug":"gpt-6-luna-xhigh","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"xhigh","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":247.25,"method":"aa_published_estimate","configurationId":"19813eb2-460a-475c-af65-810bb8660fec","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna-xhigh","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna-xhigh","sourceObservedAt":"2026-10-15T05:02:34.507Z","sourceEvidenceId":"aa-public-model-page:0e9a0c18306bc47c4296546091dc0c649d4b4a99a72f76dc332a007e4c5052cc","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["0e9a0c18306bc47c4296546091dc0c649d4b4a99a72f76dc332a007e4c5052cc"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}},{"effort":"max","label":"Max","capability":38.1,"cost":0.0678,"evidence":{"sourceId":"6fb13851-40ce-4bda-ad3f-6b38e0c5daa7","originalName":"GPT-6 Luna (Max)","sourceSlug":"gpt-6-luna","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"max","thinkingMode":"reasoning","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":435.06,"method":"aa_published_estimate","configurationId":"6fb13851-40ce-4bda-ad3f-6b38e0c5daa7","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"gpt-6-luna","sourceUrl":"https://artificialanalysis.ai/models/gpt-6-luna","sourceObservedAt":"2026-10-15T03:01:24.648Z","sourceEvidenceId":"aa-public-model-page:3e2ef7a63daebd1942c03a6b6233c1de82b01d3cb9e8605538982132617b209d","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["3e2ef7a63daebd1942c03a6b6233c1de82b01d3cb9e8605538982132617b209d"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":{"kind":"historical-reference","seconds":1284}}}]},{"id":"auto:xiaomi:c0e177b6ccb12bee","name":"MiMo-V2.6-Flash","sourceUrl":"https://artificialanalysis.ai/models/releases/mimo-v2-6-flash","recommended":"observed","chartReference":{"cost":0.0624,"capability":37.9,"z":0,"coordinateVersion":2,"benchmark":"AA Intelligence Index","capabilityEffort":"observed","costEffort":"observed","representsSingleConfiguration":true},"highCapabilityReference":{"effort":"observed","capability":37.9,"cost":0.0624,"benchmark":"AA Intelligence Index"},"recommendationReason":"AA 只测了一组未标档位的配置，按最高档推定。待核对官方文档。","recommendationPolicy":{"basis":"assumed-highest","automaticRoutingEligible":false,"comparisonPerformed":false},"configurations":[{"effort":"observed","label":"Max","capability":37.9,"cost":0.0624,"evidence":{"sourceId":"127de816-57d3-4037-8484-480a613cd413","originalName":"MiMo-V2.6-Flash","sourceSlug":"mimo-v2-6-flash","snapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","observedAt":"2026-10-15T03:01:18.201Z","nativeEffort":"observed","thinkingMode":"source-declared","fallbackPolicy":null},"generationTime":{"metric":"aa_index_generation_estimate","unit":"second","valueSeconds":1282.61,"method":"aa_published_estimate","configurationId":"127de816-57d3-4037-8484-480a613cd413","identityVerified":true,"benchmarkVersion":"v4.3.2","apiIndexVersion":"4.3","versionRelation":"same-major-minor-patch-unverified","dataSnapshotId":"aa-e797bbdb-3563-43b4-be90-1be5f2167d48","methodologyVersion":"aa-index-generation-v4","sourceRecordId":"mimo-v2-6-flash","sourceUrl":"https://artificialanalysis.ai/models/mimo-v2-6-flash","sourceObservedAt":"2026-10-15T11:32:08.192Z","sourceEvidenceId":"aa-public-model-page:e64dff5a764bc5b7dbbc2a741cd53ca7ea41c1df3fa900a0863f3d92677a0a5c","fallbackPolicy":null,"excludedComponents":["time-to-first-token","harness-setup","tool-execution","verification"],"inputEvidenceIds":["e64dff5a764bc5b7dbbc2a741cd53ca7ea41c1df3fa900a0863f3d92677a0a5c"],"displayedPrecisionSeconds":0.01},"taskTime":{"kind":"unknown","seconds":null,"range":null,"rankingEligible":false,"reference":null}}]}]}