evals.report
BenchmarksLabsCompareRun guidesIn the wild
OpenAIGPT

GPT-5 mini

OpenAI · GPT. Released Aug 7, 2025.

GPT-5 mini is a model from OpenAI in the GPT family, released Aug 7, 2025. evals.report tracks 29 reported GPT-5 mini benchmark scores across Berkeley Function Calling Leaderboard, ARC-AGI-1, SWE-bench Verified, GPQA Diamond, FrontierMath, AIME (OTIS Mock), SimpleQA Verified, ARC-AGI-2, and 21 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

29 results

Benchmark results 29

Compare this model
BenchmarkCategoryScoreMetricStatusDate
Berkeley Function Calling LeaderboardTool use55.46%accuracyOfficialAug 7, 2025Details
ARC-AGI-1Reasoning54.33%accuracyOfficialAug 7, 2025Details
SWE-bench VerifiedCoding64.7%% resolvedOfficialAug 7, 2025Details
GPQA DiamondReasoning75.0%accuracyOfficialAug 7, 2025Details
FrontierMathReasoning27.24%accuracyOfficialAug 7, 2025Details
AIME (OTIS Mock)Reasoning86.7%accuracyOfficialAug 7, 2025Details
SimpleQA VerifiedOther21.0%accuracyOfficialAug 7, 2025Details
ARC-AGI-2Reasoning4.44%accuracyOfficialAug 7, 2025Details
Humanity's Last ExamReasoning19.4%accuracyOfficialAug 7, 2025Details
WeirdMLCoding52.7%average accuracyOfficialAug 7, 2025Details
Epoch Capabilities IndexReasoning145.6IndexOfficialAug 7, 2025Details
MMLU-ProReasoning83.7%accuracyVerifiedAug 7, 2025Details
GAIA: A Benchmark for General AI AssistantsAgents44.8%accuracyUnverifiedAug 7, 2025Details
GDPvalAgents1184EloOfficialAug 7, 2025Details
LiveCodeBenchCoding83.8%Pass@1UnverifiedAug 7, 2025Details
SciCodeCoding41.0%accuracyUnverifiedAug 7, 2025Details
MultiChallengeReasoning58.99%accuracyVerifiedAug 7, 2025Details
Global-MMLUReasoning87.4%accuracyUnverifiedAug 7, 2025Details
EQ-Bench Creative Writing v3Chat preference1298EloVerifiedAug 7, 2025Details
Design ArenaChat preference1170EloVerifiedAug 7, 2025Details
MASK (Model Alignment between Statements and Knowledge)Other82.60Honesty scoreVerifiedAug 7, 2025Details
FACTS GroundingReasoning58.3%Grounding accuracyVerifiedAug 7, 2025Details
SWE-bench MultilingualCoding39.7%% resolvedOfficialAug 7, 2025Details
EnigmaEvalReasoning8.19%accuracyVerifiedAug 7, 2025Details
ZeroBenchMultimodal4.0% (pass@1)accuracyVerifiedAug 7, 2025Details
FrontierMath Tier 4Reasoning6.3%accuracyOfficialAug 7, 2025Details
Vectara Hallucination LeaderboardOther12.9%Hallucination RateOfficialAug 7, 2025Details
Vibe Code BenchCoding14.17%Overall accuracyVerifiedAug 7, 2025Details
MultiNRCReasoning23.89%accuracyOfficialAug 7, 2025Details