evals.report
BenchmarksLabsCompareRun guidesIn the wild
OpenAIGPT

GPT-5.3-Codex

OpenAI · GPT. Released Feb 5, 2026.

GPT-5.3-Codex is a model from OpenAI in the GPT family, released Feb 5, 2026. evals.report tracks 14 reported GPT-5.3-Codex benchmark scores across SWE-bench Verified, WeirdML, PostTrainBench, Artificial Analysis Intelligence Index, Epoch Capabilities Index, SWE-rebench, OSWorld, GDPval, and 6 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

14 results

Benchmark results 14

Compare this model
BenchmarkCategoryScoreMetricStatusDate
SWE-bench VerifiedCoding74.8%% resolvedOfficialFeb 5, 2026Details
WeirdMLCoding77.9%average accuracyOfficialFeb 5, 2026Details
PostTrainBenchAgents17.76%weighted average scoreOfficialFeb 5, 2026Details
Artificial Analysis Intelligence IndexReasoning53.6IndexUnverifiedFeb 5, 2026Details
Epoch Capabilities IndexReasoning155.9IndexOfficialFeb 5, 2026Details
SWE-rebenchCoding58.2%Resolved rate (pass@1)UnverifiedFeb 5, 2026Details
OSWorldAgents64.7%task success rateUnverifiedFeb 5, 2026Details
GDPvalAgents1482EloOfficialFeb 5, 2026Details
METR Task-Completion Time HorizonsAgents349.5 min50% time horizonOfficialFeb 5, 2026Details
SciCodeCoding53.2%accuracyUnverifiedFeb 5, 2026Details
WebDev ArenaChat preference1407EloVerifiedFeb 5, 2026Details
Design ArenaChat preference1199EloVerifiedFeb 5, 2026Details
Vibe Code BenchCoding61.77%Overall accuracyOfficialFeb 5, 2026Details
Terminal-Bench 2.0Agents77.3%task successVerifiedFeb 5, 2026Details