evals.report
BenchmarksLabsCompareRun guidesIn the wild

GPT-5.6 Sol

OpenAI · GPT. Released Jul 9, 2026.

GPT-5.6 Sol is a model from OpenAI in the GPT family, released Jul 9, 2026. evals.report tracks 13 reported GPT-5.6 Sol benchmark scores across AA-Omniscience: Knowledge and Hallucination Benchmark, ARC-AGI-1, ARC-AGI-2, DeepSWE, SWE-Marathon, FrontierCode, Terminal-Bench 2.1, SWE-bench Pro, and 5 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

13 results

Benchmark results 13

Compare this model
BenchmarkCategoryScoreMetricStatusDate
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning22AA-Omniscience IndexOfficialJul 9, 2026Details
ARC-AGI-1Reasoning96.5%accuracyOfficialJul 9, 2026Details
ARC-AGI-2Reasoning92.5%accuracyOfficialJul 9, 2026Details
DeepSWECoding73.0%% resolvedUnverifiedJul 9, 2026Details
SWE-MarathonAgents39.0%resolution rate (pass@1)UnverifiedJul 9, 2026Details
FrontierCodeCoding47.5%weighted score (Main)OfficialJul 9, 2026Details
Terminal-Bench 2.1Agents88.8%task successVerifiedJul 9, 2026Details
SWE-bench ProCoding64.6%% resolvedVerifiedJul 9, 2026Details
GPQA DiamondReasoning94.6%accuracyVerifiedJul 9, 2026Details
BrowseCompAgents90.4%accuracyVerifiedJul 9, 2026Details
MMMU-ProMultimodal83.0%accuracyVerifiedJul 9, 2026Details
OpenAI-MRCR v2 (Multi-Round Coreference Resolution)Reasoning73.8%accuracy (mean SequenceMatcher similarity)VerifiedJul 9, 2026Details
ARC-AGI-3Reasoning7.78%accuracyVerifiedJul 9, 2026Details