GPT-5 mini

GPT-5 mini ranks first on 3 of the 201 benchmarks with five or more models scored.

OpenAI Released August 7th, 2025 Compare with others

214

Benchmarks with a score

Ranked first, out of five or more models

In the top three

OpenAI Aug 2025

GPT-5 mini

Closed weights

146.1 #91 of 256

Epoch Capabilities Index

Self-Reported Index 142.0 #162 of 404

Humanity's Last Exam 19.4% #21 /47
SWE-bench Verified 64.7% #26 /32
MMMU-Pro 74.1% #42 /90

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
MMBench-CN v1.1	86.1%	#10 of 178	Official leaderboard
IFEval	92.7%	#6 of 60	Official leaderboard
COLLIE	98.5%	#2 of 12	Lab model card
SimpleQA Hallucination Rate	26%	#2 of 12	Lab model card
Slop Score	26.3	#3 of 17	Official leaderboard
OpenAI MRCR	84.3%	#2 of 10	Lab model card
LongFact	0.70%	#1 of 5	Lab model card
Phare	73.2%	#13 of 62	Official leaderboard
SnorkelSpatial	45.5%	#6 of 27	Official leaderboard
SuperGPQA	58.8%	#14 of 59	Official leaderboard
LLM Confabulation Leaderboard	13.3%	#14 of 58	Official leaderboard
Wer wird Millionär? LLM Benchmark	97889.0	#19 of 77	Official leaderboard
Internal API instruction following (OpenAI)	65.8%	#1 of 4	Lab model card
LongFact-Concepts	0.90%	#1 of 4	Lab model card
Fiction.LiveBench	69.4%	#19 of 57	Fiction.live leaderboard, via Epoch AI
Vectara Hallucination Leaderboard (HHEM)	3.20%	#50 of 148	Official leaderboard
MultiChallenge	59%	#17 of 42	Official leaderboard
LMArena (Elo)	1,390	#143 of 335	Official leaderboard
Humanity's Last Exam	19.4%	#21 of 47	Official leaderboard
LongFact-Objects	1.80%	#2 of 4	Lab model card
FACTS Search	67.9%	#16 of 31	Official leaderboard
GraphWalks	73.4%	#16 of 27	Lab model card
SimpleQA	22%	#37 of 60	Lab model card
Longform Creative Writing (EQ-Bench)	56.2	#75 of 119	Official leaderboard
MultiNRC	23.9%	#25 of 38	Official leaderboard
BrowseComp Long Context 128k	89.4%	#2 of 3	Lab model card
FActScore	3.50%	#2 of 3	Lab model card
Creative Writing v3	60.0	#84 of 120	Official leaderboard
Structured Output Benchmark	83.5%	#14 of 19	Official leaderboard
HUMAINE	22	#38 of 50	Official leaderboard
FACTS Parametric	16.1%	#35 of 45	Official leaderboard
SkateBench v2	38%	#17 of 21	Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset)	12.9%	#86 of 104	Official leaderboard
SimpleQA Verified	21.6%	#79 of 94	Epoch AI run
HealthBench Hard Hallucinations	3.80%	#2 of 2	Lab model card

Math and science

Benchmark	Score	Rank	Source
MATH Level 5	97.8%	#2 of 94	Epoch AI run
GPQA Main	75.6%	#3 of 60	Official leaderboard
SnorkelSequences	77.6%	#2 of 25	Official leaderboard
IneqMath	30.5%	#4 of 34	Official leaderboard
SnorkelGraph	72.5%	#4 of 24	Official leaderboard
MATH-500	94.8%	#7 of 38	Official leaderboard
SciArena	1,067	#7 of 34	Official leaderboard
CMT-Benchmark	24%	#3 of 11	Official leaderboard
SMT 2025	89.2%	#11 of 40	Official leaderboard
AIME 2025	91.1%	#44 of 143	Lab model card
FrontierMath	27.2%	#18 of 55	Official leaderboard
OTIS Mock AIME 2024-2025	86.7%	#79 of 212	Epoch AI run
FrontierMath Tier 4	6.25%	#24 of 55	Epoch AI run
CMIMC 2025	84.4%	#15 of 33	Official leaderboard
DTBench	80.5%	#88 of 180	Official leaderboard
GPQA Diamond	75%	#112 of 229	Epoch AI run
HMMT 2025	87.8%	#27 of 54	Lab model card
BRUMO 2025	90%	#24 of 42	Official leaderboard
ForecastBench	0.14	#48 of 82	Official leaderboard
FrontierMath Tiers 1-3 v2	46.7%	#54 of 86	Official leaderboard
ARC-AGI	54.3%	#65 of 95	Official leaderboard
ARC-AGI-2	4.44%	#72 of 98	Official leaderboard
ProofBench	9.00%	#45 of 59	Official leaderboard
AstaBench	31.4%	#10 of 13	Official leaderboard
HMMT Nov 2025	84.2%	#17 of 22	Official leaderboard
FrontierMath Tier 4 v2	12.2%	#54 of 68	Official leaderboard
MathArena (Apex, IMO, USAMO)	1.04%	#39 of 47	Official leaderboard
MathArena Apex Shortlist	39.9%	#34 of 38	Official leaderboard
A^3-Bench	25.3%	#7 of 7	Official leaderboard

Coding

Benchmark	Score	Rank	Source
SWT-bench	56.2%	#2 of 7	Official leaderboard
Aider Polyglot	71.6%	#10 of 30	Lab model card
MMTU	66.7%	#3 of 9	Official leaderboard
WeirdML	52.7%	#43 of 112	Official leaderboard
ProgramBench	0%	#11 of 25	Official leaderboard
Spider 2.0	41%	#11 of 23	Official leaderboard
ALE-Bench	1,421	#66 of 126	Official leaderboard
OpenAI PRs	42%	#7 of 12	Lab model card
CodeClash	1,199	#5 of 8	Official leaderboard
IOI (Vals)	6.75%	#36 of 55	Official leaderboard
SWE-Lancer	51%	#7 of 10	Lab model card
Vibe Code Bench v1	4.87%	#12 of 17	Official leaderboard
ProgramBench almost solved	0%	#20 of 25	Official leaderboard
PaperBench	23%	#12 of 15	Lab model card
Vibe Code Bench	14.2%	#97 of 120	Official leaderboard
SWE-bench Verified	64.7%	#26 of 32	Epoch AI run
AlgoTune	1.38×	#15 of 18	Official leaderboard
DAXBench v2	63%	#41 of 47	Official leaderboard
SWE-fficiency	3.89%	#15 of 17	Official leaderboard
Data Agent Benchmark	36.6%	#17 of 18	Official leaderboard
ProgramBench tests passed	16%	#25 of 25	Official leaderboard
SWE-bench Multilingual	39.7%	#13 of 13	Official leaderboard
OpenAI-Proof Q&A	1.00%	#9 of 9	Lab model card
MLE-bench 30	3.00%	#8 of 8	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
Elimination Game	6	#3 of 56	Official leaderboard
TextQuests	12.0	#4 of 28	Official leaderboard
LLM-WikiRace	11%	#4 of 21	Official leaderboard
AdaPlanBench	61.9%	#2 of 10	Official leaderboard
BFCL V4	55.5%	#14 of 65	Official leaderboard
SnorkelWordle	91%	#5 of 19	Official leaderboard
LLM Chess	759	#34 of 123	Official leaderboard
MCPMark	30.3%	#8 of 28	Official leaderboard
LiveResearchBench	67.7	#2 of 7	Official leaderboard
Terminal-Bench	34.8%	#6 of 18	tbench.ai, via Epoch AI
LisanBench	1561.7	#48 of 133	Official leaderboard
SnakeBench	26	#73 of 188	Official leaderboard
CompileBench	80%	#8 of 20	Official leaderboard
Step Race	3	#24 of 59	Official leaderboard
Chess Puzzles	24%	#58 of 93	Epoch AI run
Game Arena Chess Openings	484	#7 of 11	Official leaderboard
Game Arena Five in a Row	387	#13 of 20	Official leaderboard
Terminal-Bench 2.0	34.8%	#29 of 43	Official leaderboard
Game Arena Dark Hex	188	#28 of 41	Official leaderboard
PinchBench	65.2%	#33 of 48	Official leaderboard
Game Arena Dots and Boxes	708	#29 of 40	Official leaderboard
Game Arena Checkers	683	#30 of 41	Official leaderboard
tau2-bench	70.8%	#58 of 79	Lab model card
Game Arena Bargaining	143	#31 of 39	Official leaderboard
Game Arena Reversi	202	#31 of 39	Official leaderboard
Game Arena Word Association	83	#32 of 40	Official leaderboard
Clocktower Radio	1,393	#19 of 23	Official leaderboard
Mystery Game Puzzles	10%	#70 of 84	Epoch AI run
Game Arena Chess	539	#35 of 42	Official leaderboard
EnterpriseOps-Gym	22.9%	#10 of 12	Official leaderboard
Game Arena Coin Game	228	#33 of 39	Official leaderboard
Game Arena Nine Men's Morris	292	#34 of 39	Official leaderboard
Game Arena Clobber	43	#35 of 39	Official leaderboard
Kaggle Game Arena	159	#38 of 42	Official leaderboard
tau2-bench Verified	69.8%	#10 of 11	Official leaderboard
Toolathlon	14.5%	#46 of 50	Official leaderboard
Game Arena Lines of Action	138	#37 of 39	Official leaderboard
Game Arena Ultimate Tic-Tac-Toe	17	#38 of 39	Official leaderboard
Game Arena Four in a Row	0	#43 of 43	Official leaderboard
Game Arena Go	0	#39 of 39	Official leaderboard
Game Arena Word Art	0	#39 of 39	Official leaderboard
Game Arena Poker	-49.68	#31 of 31	Official leaderboard
Learn2Play Bench	53.5	#11 of 11	Official leaderboard
OpenThoughts-TBLite	50.5%	#4 of 4	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
CaseLaw v2	68.5%	#4 of 53	Official leaderboard
HealthBench Consensus	96.5%	#2 of 17	Lab model card
HealthBench	64.1%	#4 of 31	Lab model card
LEXam	60.3%	#4 of 30	Official leaderboard
TaxEval v2	75.2%	#22 of 129	Official leaderboard
FinQA (Snorkel OpenEnv environment)	79.3%	#5 of 22	Official leaderboard
SnorkelFinance	79.3%	#5 of 22	Official leaderboard
MedHELM	69.0	#3 of 13	Official leaderboard
HealthBench Hard	40.3%	#6 of 21	Lab model card
MortgageTax	66.9%	#29 of 87	Official leaderboard
Finance Reasoning (Snorkel)	46.8%	#8 of 21	Official leaderboard
FailureSensorIQ	65.3%	#10 of 24	Official leaderboard
SnorkelUnderwrite	71.7%	#12 of 27	Official leaderboard
MedCode	43%	#46 of 96	Official leaderboard
Vals Finance Agent	51.9%	#26 of 49	Official leaderboard
MedScribe	80.6%	#52 of 97	Official leaderboard
LegalBench	81.8%	#77 of 138	Official leaderboard
CorpFin v2	60.2%	#69 of 119	Official leaderboard
ProfBench	41.6%	#58 of 72	Official leaderboard
RCM-Bench-1	84.3%	#17 of 20	Official leaderboard
Vals Index v1	44.5%	#29 of 34	Official leaderboard
FoodTruck Bench	$50	#55 of 59	Official leaderboard
Vending-Bench 2	$-31	#67 of 67	Official leaderboard
Vending-Bench Arena	$154	#32 of 32	Official leaderboard
HealthBench Consensus Global Health	0.50%	#2 of 2	Lab model card
HealthBench Consensus Urgent	1.40%	#2 of 2	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
MMStar	73.2%	#4 of 173	Official leaderboard
MMBench-EN v1.1	86.2%	#11 of 178	Official leaderboard
MM-Vet	74.6%	#12 of 164	Official leaderboard
RoboRewardBench	0.69	#2 of 22	Official leaderboard
Progress-Bench	20.9	#1 of 9	Official leaderboard
MMMU	81.6%	#16 of 103	Lab model card
VISTA	50.4%	#10 of 54	Official leaderboard
MeasureBench Real-world	22%	#4 of 19	Official leaderboard
MATH-Vision (MathVision)	71.9%	#24 of 105	Official leaderboard
MeasureBench Synthetic	17.9%	#5 of 19	Official leaderboard
ChartMuseum	63.3%	#6 of 21	Official leaderboard
Blueprint-Bench	0.40	#3 of 8	Official leaderboard
ERQA	62.9%	#16 of 36	Lab model card
MADQA	66.9%	#8 of 18	Official leaderboard
MMMU-Pro	74.1%	#42 of 90	Lab model card
Video-MMMU	82.5%	#17 of 36	Lab model card
VPCT	40.2%	#14 of 27	VPCT leaderboard, via Epoch AI
ZeroBench	9.00%	#40 of 76	Official leaderboard
VisFactor	24.3%	#14 of 26	Official leaderboard
Math Kangaroo 2025 (MathArena)	78.2%	#13 of 22	Official leaderboard
Video-MME	78.5%	#32 of 51	Lab model card
Vision Arena	1,185	#87 of 136	Official leaderboard
CharXiv	75.5%	#48 of 75	Lab model card
ParseBench	49.9%	#33 of 44	Official leaderboard
RoboSpatial-Home	51.3%	#3 of 4	Official leaderboard
MineBench	1,076	#63 of 75	Official leaderboard
ClockBench	8.90%	#32 of 37	Official leaderboard
vCX Benchmark	75%	#38 of 42	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
HELM Safety	98.0	#6 of 75	Official leaderboard
AIR-Bench 2024	85.7%	#13 of 82	Official leaderboard
SimpleSafetyTests	100%	#20 of 75	Official leaderboard
FORTRESS	17%	#22 of 73	Official leaderboard
Capture the Flag (Collegiate)	36%	#4 of 13	Lab model card
Challenging prompts from red teamers with biosafety-relevant PhDs	93.6%	#1 of 3	Lab model card
Molecular Biology Capabilities Test	51.5%	#34 of 90	Official leaderboard
World-Class Bio	49.2%	#30 of 74	Official leaderboard
Human Pathogen Capabilities Test	54.2%	#38 of 86	Official leaderboard
ProtocolQA Open-Ended	37%	#12 of 27	Lab model card
Anthropic Red Team	99.1%	#34 of 75	Official leaderboard
KORA	43.2%	#21 of 42	Official leaderboard
Capture the Flag (Professional)	23%	#13 of 21	Lab model card
Filtered, adversarial sample of production prompts	96.8%	#2 of 3	Lab model card
Tacit Knowledge and Troubleshooting	68%	#22 of 30	Lab model card
BrowseSafe-Bench	85.4%	#3 of 4	Official leaderboard
TroubleshootingBench	26%	#17 of 20	Lab model card
nocot-bench (NCRI)	50.7	#162 of 185	Official leaderboard
Multimodal Troubleshooting Virology	39%	#26 of 27	Lab model card

Other

Benchmark	Score	Rank	Source
HELM Capabilities	81.9	#1 of 60	Official leaderboard
OpenVLM Leaderboard	78.0	#4 of 173	Official leaderboard
ACCoRD	71.5	#42 of 152	Official leaderboard
Conceptual Reasoning Index	48.3	#76 of 151	Official leaderboard
LMCA	34.2%	#81 of 147	Official leaderboard
CAIS Vision Capabilities Index	47.5	#32 of 53	Official leaderboard
SAGE (Vals)	43%	#52 of 81	Official leaderboard
CAIS Risk Index	58.5	#48 of 67	Official leaderboard
CAIS Text Capabilities Index	18.5	#54 of 68	Official leaderboard

Its lab's numbers put it at 142.0, 4.1 points below its ECI

The Self-Reported Index fits one score to the 26 capability results OpenAI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 139.5 to 144.6. How we compute it.

Its lab's numbers next to 7 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
SWE-bench Verified	72% SWE-bench Verified (n=477), pass@1, gpt-5-thinking-mini no browsing launch candidate (Figure 19)	59.8% Leaderboard, medium, mini-SWE-agent	+12 pts
GPQA Diamond	82.3% GPQA diamond (no tools)	75% Epoch AI run, high	+7.3 pts
FrontierMath	22.1% FrontierMath (with python tool only)	27.2% Leaderboard, high; task version 1.1.0, Epoch	−5.1 pts
AIME 2025	91.1% AIME '25 (no tools)	87.5% Leaderboard, high	+3.6 pts
MultiChallenge	62.3% Scale MultiChallenge (o3-mini grader)	59% Leaderboard, thinking	+3.3 pts
Humanity's Last Exam	16.7% HLE (no tools)	19.4% Leaderboard	−2.7 pts
HMMT 2025	87.8% HMMT 2025 (no tools)	89.2% Leaderboard, high	−1.4 pts

OpenAI's models over time

OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Introducing GPT-5 for developers (OpenAI) Aug 7, 2025

Its system card

Every result in the longer documents the lab published with it. 28 of them aren't in the launch post.

GPT-5 mini

GPT-5 mini

Where it ranks

Knowledge and language

Math and science

Coding

Agents and tools

Professional work

Vision, audio and robotics

Safety and security

Other

Its lab's numbers put it at 142.0, 4.1 points below its ECI

Its lab's numbers next to 7 leaderboards

OpenAI's models over time

Its launch post

Its system card

Health

Hallucination

Safety

Robustness

Bio and chem risk

Cyber risk

AI R&D risk