ARC-AGI-3 is an interactive reasoning benchmark in which an agent learns unfamiliar game-like environments through actions on first exposure. AgMoDB records the official 25-environment Public Demo score with its relative-human-action-efficiency metric and action-budget policy.
Interactive exploration and execution test online model building beyond static question answering; action-efficiency scoring penalizes brute force and compares completed levels with first-run human baselines.
The current public records publish model variants and total evaluation cost but not exact provider model IDs, harness versions, or upstream run IDs. AgMoDB preserves those gaps as null and blocks the rows from autonomous recommendation changes until comparable provenance is published.
Higher is better. Percentage relative-human-action-efficiency across 25 Public Demo environments, with each level capped at five times the human-baseline median action count.
Max score: 100
The current public records publish model variants and total evaluation cost but not exact provider model IDs, harness versions, or upstream run IDs.
| # | Model | Score |
|---|---|---|
| 1 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (XHigh) · xhigh · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-xhigh-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "openai-gpt-6-2-astra-xhigh-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4112.399250400001,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 100.0 |
| 2 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (XHigh) · xhigh · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-xhigh-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"d8dfdc9958a2e1502b54603b03fbc557\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"aaa10807f3dcaae3ce6c15be7539e733\"",
"upstreamModelId": "openai-gpt-6-2-astra-xhigh-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4112.399250400001,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 100.0 |
| 3 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (XHigh) · xhigh · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-xhigh-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-xhigh-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4112.399250400001,
"providerModelIdPublished": false
}Evidence observed Nov 18, 2026 | 100.0 |
| 4 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (XHigh) · xhigh · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-xhigh-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'xhigh' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-xhigh-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4112.399250400001,
"providerModelIdPublished": false
}Evidence observed Nov 17, 2026 | 100.0 |
| 5 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.9 |
| 6 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"d8dfdc9958a2e1502b54603b03fbc557\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"aaa10807f3dcaae3ce6c15be7539e733\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.9 |
| 7 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 18, 2026 | 99.9 |
| 8 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 17, 2026 | 99.9 |
| 9 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"69aae253e532553a920bf4deb50ca495\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5206202f381a21893de82dcc2db93ae9\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 16, 2026 | 99.9 |
| 10 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"513aaaa353972ca0baa9c95cb95e04dd\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"169653a17d103c8dbca682167ffccaa6\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 10, 2026 | 99.9 |
| 11 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"3d7dcca4fb88130b9e7ec186fd9ae23f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"1e46e6ab8c83fb1a7b19f10be398adf2\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 9, 2026 | 99.9 |
| 12 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"8ea4d2292a0a8c759aa7e80a2041122d\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"94fd3d68aacca744ff3c39f63a187685\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 6, 2026 | 99.9 |
| 13 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"c4baed683b2f33c40537f2338682475f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"48d1fd5bf96ae9583213f6699f12f2cc\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 4, 2026 | 99.9 |
| 14 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"c4baed683b2f33c40537f2338682475f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"48d1fd5bf96ae9583213f6699f12f2cc\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Nov 2, 2026 | 99.9 |
| 15 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"8fc20a5217998d26c40ad2306d648194\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"095b2fa20aac96121092dab07e18cc8e\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 28, 2026 | 99.9 |
| 16 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"400b32b66e8c1b37fc50ff586b033048\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"9314d4bea0d9d001c69a11614dc8d93c\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 21, 2026 | 99.9 |
| 17 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"2ab08f641c16bcf3c7a59c38b35fa2e3\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"b4ed41aee6b00420557f098bd05f4165\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 19, 2026 | 99.9 |
| 18 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"8cf5cb76df0b13c1684b346a2012df83\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"c6e9a78d12dc070d7dab18a1f667a410\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 15, 2026 | 99.9 |
| 19 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"57cc4a7689e563a0ffb7d275eb3a889b\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"19d71f8c9a2056c0be4bb4fd4a75a325\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 14, 2026 | 99.9 |
| 20 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"57cc4a7689e563a0ffb7d275eb3a889b\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"19d71f8c9a2056c0be4bb4fd4a75a325\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 14, 2026 | 99.9 |
| 21 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"e7f673a6db195dac7f2956e3b1651145\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"4d3b46c6ffe9404ca1c0ebcfb5938883\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 13, 2026 | 99.9 |
| 22 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"b57a26ca4df83acaa40eadb6632c7c61\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ee35c9dbbb033dc7de85467eedb85d54\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 6, 2026 | 99.9 |
| 23 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"b52a6e82589fb2c719168d69aa0bb015\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"f187ceaff4d7f6156e77c1f1254aafc8\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 2, 2026 | 99.9 |
| 24 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"b52a6e82589fb2c719168d69aa0bb015\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"f187ceaff4d7f6156e77c1f1254aafc8\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Oct 1, 2026 | 99.9 |
| 25 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"8bfe2e142f29acec39df406a6631f81a\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"92bfaa47028eae49163960f4c5a298a0\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 30, 2026 | 99.9 |
| 26 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"16793624264c3bdded076b4e26af5a90\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"e7fb37556a7353e7da435c4c7b78ca50\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 29, 2026 | 99.9 |
| 27 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"284ac3aff89a7d8a3aca179ee1e5dd61\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"6e0849a05a2c572b1050e5900a372e2b\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 28, 2026 | 99.9 |
| 28 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"397d7634d348db2ae8df5180db7b475d\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"882404aedf04dbdde2b6c0d731212a4f\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 24, 2026 | 99.9 |
| 29 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"58c953e3f78e2282d0259b52304c8ef8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef004f05f102510151398f73e966a509\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 22, 2026 | 99.9 |
| 30 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"2f3beadbd513a221b41d3816ff651f54\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"f1d0d3b7e367e4a7909882624b1caca0\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 22, 2026 | 99.9 |
| 31 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"c985e4804ffc840374d9852c6b60005c\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"0a820529c69911b0c1a63d8b2d3dbf0f\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 21, 2026 | 99.9 |
| 32 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"a8dc9c600e07efa11ca8fb622181783f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"946807139d47f4bc368591b95885dfb4\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 4, 2026 | 99.9 |
| 33 | GPT-6 Astra (High) source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"a8dc9c600e07efa11ca8fb622181783f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"946807139d47f4bc368591b95885dfb4\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 4, 2026 | 99.9 |
| 34 | OpenAI: GPT-6 Astra Pro source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"a8dc9c600e07efa11ca8fb622181783f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"946807139d47f4bc368591b95885dfb4\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 4, 2026 | 99.9 |
| 35 | GPT-6 Astra (High) source arc_agi · GPT-6 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"a8dc9c600e07efa11ca8fb622181783f\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"946807139d47f4bc368591b95885dfb4\"",
"upstreamModelId": "openai-gpt-6-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 18816.63138,
"providerModelIdPublished": false
}Evidence observed Sep 3, 2026 | 99.9 |
| 36 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (Max) · max · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-max-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "openai-gpt-6-2-astra-max-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4106.9577648,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.8 |
| 37 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (Max) · max · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-max-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"d8dfdc9958a2e1502b54603b03fbc557\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"aaa10807f3dcaae3ce6c15be7539e733\"",
"upstreamModelId": "openai-gpt-6-2-astra-max-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4106.9577648,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.8 |
| 38 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (Max) · max · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-max-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-max-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4106.9577648,
"providerModelIdPublished": false
}Evidence observed Nov 18, 2026 | 99.8 |
| 39 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (Max) · max · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-max-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-max-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4106.9577648,
"providerModelIdPublished": false
}Evidence observed Nov 17, 2026 | 99.8 |
| 40 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "openai-gpt-6-2-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 3878.2227588000005,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.8 |
| 41 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs.",
"modelsEtag": "W/\"d8dfdc9958a2e1502b54603b03fbc557\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"aaa10807f3dcaae3ce6c15be7539e733\"",
"upstreamModelId": "openai-gpt-6-2-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 3878.2227588000005,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.8 |
| 42 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 3878.2227588000005,
"providerModelIdPublished": false
}Evidence observed Nov 18, 2026 | 99.8 |
| 43 | OpenAI: GPT-6.2 Astra Pro source arc_agi · GPT-6.2 Astra - Provider Adapter (High) · high · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-2-astra-high-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-2-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'high' reasoning effort using the provider adapter harness. Some model actions have unmeasured token usage; reported costs are recorded-usage lower bounds, not verified full-run costs.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "openai-gpt-6-2-astra-high-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-17T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 3878.2227588000005,
"providerModelIdPublished": false
}Evidence observed Nov 17, 2026 | 99.8 |
| 44 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.1 |
| 45 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"d8dfdc9958a2e1502b54603b03fbc557\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"aaa10807f3dcaae3ce6c15be7539e733\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 99.1 |
| 46 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 18, 2026 | 99.1 |
| 47 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"f4d8307b32b2fefe010860bedd0c5dd8\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"ef15ebd7644a875688eda094d716da9a\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 17, 2026 | 99.1 |
| 48 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"69aae253e532553a920bf4deb50ca495\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5206202f381a21893de82dcc2db93ae9\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 16, 2026 | 99.1 |
| 49 | Gemini 4 Carbon (Medium) source arc_agi · Gemini 4 Carbon - Provider Adapter (Medium) · medium · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:google-gemini-4-carbon-medium-provider-adapter View run sourceSource metadata{
"provider": "Google",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "google-gemini-4-carbon-provider-adapter",
"modelNotes": "Evaluation conducted with 'medium' reasoning effort using the provider adapter harness. Partial dataset coverage: scores use full catalog denominators; costs include recorded usage only, excluding unrecorded attempts.",
"modelsEtag": "W/\"513aaaa353972ca0baa9c95cb95e04dd\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"169653a17d103c8dbca682167ffccaa6\"",
"upstreamModelId": "google-gemini-4-carbon-medium-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-11-10T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 4559.504952,
"providerModelIdPublished": false
}Evidence observed Nov 10, 2026 | 99.1 |
| 50 | GPT-6 Astra source arc_agi · GPT-6 Astra - Provider Adapter (Max) · max · benchmark 3 · split public-demo · 1 attempts · 5x-human-baseline-median-per-level · arc-agi-3:public-demo · record v3_Semi_Private:openai-gpt-6-astra-max-provider-adapter View run sourceSource metadata{
"provider": "OpenAI",
"modelType": "CoT",
"modelsUrl": "https://arcprize.org/media/data/models.json",
"modelGroup": "openai-gpt-6-astra-provider-adapter",
"modelNotes": "Evaluation conducted with 'max' reasoning effort using the provider adapter harness.",
"modelsEtag": "W/\"ec09fbed9f211f59ec54cca2227c3309\"",
"scoreMetric": "relative-human-action-efficiency",
"evaluationsUrl": "https://arcprize.org/media/data/evaluations.json",
"evaluationsEtag": "W/\"5814b0938ba4a54ba31da46fe17d4380\"",
"upstreamModelId": "openai-gpt-6-astra-max-provider-adapter",
"environmentCount": 25,
"modelReleaseDate": "2026-09-02T00:00:00.000Z",
"upstreamDatasetId": "v3_Semi_Private",
"actionBudgetPolicy": "5x-human-baseline-median-per-level",
"technicalReportUrl": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
"catalogMatchAliases": [],
"agentHarnessPublished": false,
"verificationAuthority": "ARC Prize Foundation",
"upstreamRunIdPublished": false,
"actualEvaluationCostUsd": 17331.95327,
"providerModelIdPublished": false
}Evidence observed Nov 20, 2026 | 98.6 |