evals.report
BenchmarksLabsCompareRun guidesIn the wild

GPT-6 Astra

OpenAI · GPT. Released Sep 3, 2026.

GPT-6 Astra is a model from OpenAI in the GPT family, released Sep 3, 2026. evals.report tracks 9 reported GPT-6 Astra benchmark scores across FrontierCode, AA-Omniscience: Knowledge and Hallucination Benchmark, ARC-AGI-1, ARC-AGI-2, GPQA Diamond, BrowseComp, OpenAI-MRCR v2 (Multi-Round Coreference Resolution), DeepSWE, and 1 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

9 results

Benchmark results 9

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierCodeCoding53.3%weighted score (Main)OfficialSep 3, 2026Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning44AA-Omniscience IndexOfficialSep 3, 2026Details
ARC-AGI-1Reasoning97.5%accuracyOfficialSep 3, 2026Details
ARC-AGI-2Reasoning95%accuracyOfficialSep 3, 2026Details
GPQA DiamondReasoning96%accuracyVerifiedSep 3, 2026Details
BrowseCompAgents91.5%accuracyVerifiedSep 3, 2026Details
OpenAI-MRCR v2 (Multi-Round Coreference Resolution)Reasoning96.3%accuracy (mean SequenceMatcher similarity)VerifiedSep 3, 2026Details
DeepSWECoding74.1%% resolvedVerifiedSep 3, 2026Details
ScreenSpot-ProMultimodal92.7%accuracyVerifiedSep 3, 2026Details