Benchmark Almanac

GPT-5

GPT-5 ranks first on 52 of the 296 benchmarks with five or more models scored.

OpenAI Released August 7th, 2025 Family GPT-5 The lab's model card Compare with others
325
Benchmarks with a score
52
Ranked first, out of five or more models
96
In the top three
OpenAI Aug 2025

GPT-5

Closed weights Flagship

150.0 #70 of 256
Epoch Capabilities Index
Self-Reported Index 147.5 #114 of 404
  • Humanity's Last Exam 25.3% #15 /47
  • SWE-bench Verified 73.5% #19 /32
  • SWE-Bench Pro 41.8% #10 /24

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

BenchmarkScoreRankSource
LLM Confabulation Leaderboard 10.3% #1 of 58 Official leaderboard
WritingBench 83.9 #1 of 45 Official leaderboard
Fiction.LiveBench 97.2% #2 of 57 Fiction.live leaderboard, via Epoch AI
MMBench-CN v1.1 86.4% #8 of 178 Official leaderboard
AraGen 84.3% #2 of 42 Official leaderboard
FACTS Grounding 87.4% #2 of 34 Official leaderboard
Wer wird Millionär? LLM Benchmark 366667.0 #5 of 77 Official leaderboard
ConStory-Bench 0.11 #1 of 14 Official leaderboard
MMLU-Pro 87.1% #16 of 212 Official leaderboard
Vectara Hallucination Leaderboard (HHEM) 1.40% #12 of 148 Official leaderboard
COLLIE 99% #1 of 12 Lab model card
SuperGPQA 65.6% #5 of 59 Official leaderboard
DiscoX 76.7 #1 of 7 Official leaderboard
SnorkelSpatial 73.9% #4 of 27 Official leaderboard
SimpleQA 55% #10 of 60 Lab model card
SkateBench v2 77.4% #4 of 21 Official leaderboard
FACTS Search 77.7% #6 of 31 Official leaderboard
TableBench 59.9% #2 of 10 Official leaderboard
MultiNRC 52.1% #9 of 38 Official leaderboard
LMArena (Elo) 1,435 #83 of 335 Official leaderboard
IFEval 87.5% #15 of 60 Official leaderboard
LongFact-Objects 1.40% #1 of 4 Lab model card
Longform Creative Writing (EQ-Bench) 72.8 #34 of 119 Official leaderboard
MultiChallenge 63.2% #13 of 42 Official leaderboard
Creative Writing v3 72.7 #38 of 120 Official leaderboard
Humanity's Last Exam 25.3% #15 of 47 Official leaderboard
SimpleQA Verified 50.1% #31 of 94 Epoch AI run
FActScore 1.00% #1 of 3 Lab model card
Visual Aesthetic Benchmark 32.3% #7 of 18 Official leaderboard
FACTS Parametric 55.8% #18 of 45 Official leaderboard
LongFact 0.80% #2 of 5 Lab model card
SimpleQA Hallucination Rate 40% #5 of 12 Lab model card
Phare 67.2% #28 of 62 Official leaderboard
IndicGenBench 52.2 #6 of 13 Official leaderboard
HealthBench Hard Hallucinations 1.60% #1 of 2 Lab model card
Joke Arena 1,506 #13 of 25 Official leaderboard
Global MMLU Lite 89% #23 of 43 Official leaderboard
BullshitBench 21.8% #66 of 107 Official leaderboard
Structured Output Benchmark 84.9% #12 of 19 Official leaderboard
BullshitBench V2 21% #90 of 129 Official leaderboard
DesignArena ASCII Art 1,136 #55 of 77 Official leaderboard
LongFact-Concepts 1.10% #3 of 4 Lab model card
HUMAINE 22 #39 of 50 Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset) 14.7% #94 of 104 Official leaderboard
Corp Speak Arena 1,367 #20 of 22 Official leaderboard
Opening Lines Arena 1,328 #15 of 15 Official leaderboard
AbstentionBench 53% #2 of 2 Lab model card
Hallucinations on production-representative prompts 4.50% #2 of 2 Lab model card

Math and science

BenchmarkScoreRankSource
MATH Level 5 98.1% #1 of 94 Epoch AI run
AIME 2025 100% #2 of 143 Lab model card
HMMT 2025 100% #1 of 54 Lab model card
IneqMath 47% #1 of 34 Official leaderboard
GPQA Main 79.2% #2 of 60 Official leaderboard
SnorkelSequences 77.6% #1 of 25 Official leaderboard
AMO-Bench 52.4% #1 of 21 Official leaderboard
SMT 2025 92% #3 of 40 Official leaderboard
MATH-500 96% #3 of 38 Official leaderboard
SciArena 1,123 #3 of 34 Official leaderboard
CMT-Benchmark 30% #1 of 11 Official leaderboard
HiPhO 67.9% #4 of 36 Official leaderboard
PHYBench 40.4% #4 of 28 Official leaderboard
IMO 2025 38.1% #1 of 6 Official leaderboard
General365 58.6% #4 of 23 Official leaderboard
ForecastBench 0.12 #15 of 82 Official leaderboard
ATLAS (OpenCompass) 42.9% #1 of 5 Official leaderboard
SnorkelGraph 72% #5 of 24 Official leaderboard
DTBench 90.7% #44 of 180 Official leaderboard
FrontierMath 32.4% #14 of 55 Official leaderboard
OTIS Mock AIME 2024-2025 91.4% #63 of 212 Epoch AI run
CMIMC 2025 88.8% #10 of 33 Official leaderboard
FrontierMath Tier 4 12.5% #17 of 55 Epoch AI run
GPQA Diamond 86.2% #75 of 229 Epoch AI run
SGI-Bench 30.8% #6 of 15 Official leaderboard
LiveMathematicianBench 28.3% #7 of 17 Official leaderboard
SimpleBench 56.7% #50 of 106 Official leaderboard
BRUMO 2025 91.7% #20 of 42 Official leaderboard
CORE-Bench 26.7% #9 of 17 Official leaderboard
AstaBench 43.3% #7 of 13 Official leaderboard
IMO-ProofBench 20% #11 of 20 Official leaderboard
ProofBench 18% #33 of 59 Official leaderboard
FrontierMath Tiers 1-3 v2 55.4% #49 of 86 Official leaderboard
ARC-AGI 65.7% #56 of 95 Official leaderboard
HMMT Nov 2025 89.2% #13 of 22 Official leaderboard
ARC-AGI-2 9.86% #60 of 98 Official leaderboard
MathArena (Apex, IMO, USAMO) 2.08% #31 of 47 Official leaderboard
FrontierMath Tier 4 v2 21.9% #45 of 68 Official leaderboard
IMProofBench 20.4% #6 of 9 Official leaderboard
CritPt 28.9% #11 of 11 Official leaderboard
CritPt v2 28.9% #11 of 11 Official leaderboard

Coding

BenchmarkScoreRankSource
Aider Polyglot 88% #1 of 30 Lab model card
ArtifactsBench 72.5% #1 of 20 Official leaderboard
METR Time Horizon 2.2 h #3 of 31 Official leaderboard
ContextBench 60.6% #1 of 9 Official leaderboard
MMTU 69.6% #1 of 9 Official leaderboard
ts-bench v1 96% #1 of 9 Official leaderboard
SWE-fficiency 15.7% #2 of 17 Official leaderboard
CooperBench 27.9% #1 of 6 Official leaderboard
Spider 2.0 65.6% #4 of 23 Official leaderboard
CATArena Gomoku 1,799 #1 of 5 Official leaderboard
CATArena OIBench-5 0.40 #1 of 5 Official leaderboard
CATArena Texas Holdem 1,463 #1 of 5 Official leaderboard
CodeClash 1,366 #2 of 8 Official leaderboard
WeirdML 60.7% #32 of 112 Official leaderboard
SWE-Lancer 55% #3 of 10 Lab model card
OpenAI PRs 45% #4 of 12 Lab model card
BIRD-Interact 25.5% #1 of 3 Official leaderboard
METR Time Horizon 1.1 3.4 h #10 of 26 Official leaderboard
CATArena SWE-Perf-10 1.14× #2 of 5 Official leaderboard
Vision2Web S1-2026 40.5% #2 of 5 Official leaderboard
ALE-Bench 1,545 #52 of 126 Official leaderboard
SWE-Bench Pro 41.8% #10 of 24 Official leaderboard
IOI (Vals) 20% #23 of 55 Official leaderboard
ABC-Bench 49.4% #3 of 7 Official leaderboard
LiveSQLBench 25.5% #16 of 36 Official leaderboard
AlgoTune 1.67× #8 of 18 Official leaderboard
DesignArena SVG 1,179 #48 of 103 Official leaderboard
Vibe Code Bench v1 20.1% #8 of 17 Official leaderboard
DesignArena Data Visualization 1,219 #57 of 118 Official leaderboard
MLE-bench 39.6% #6 of 12 Official leaderboard
AInsteinBench 35.4% #3 of 6 Official leaderboard
FlashInfer-Bench 0.47× #1 of 2 Official leaderboard
DesignArena Game Dev 1,175 #76 of 143 Official leaderboard
Snorkel Agentic Coding 45.2% #6 of 11 Official leaderboard
DesignArena UI Component 1,171 #80 of 139 Official leaderboard
DesignArena Website 1,177 #88 of 151 Official leaderboard
SWE-bench Verified 73.5% #19 of 32 Epoch AI run
WebDev Arena 1,420 #63 of 105 arena.ai, via Epoch AI
Code Arena 1,417 #80 of 128 Official leaderboard
DesignArena Code 1,158 #91 of 143 Official leaderboard
GSO-Bench 6.86% #20 of 30 Official leaderboard
SWE-bench-Live 24.3% #8 of 12 Official leaderboard
Vibe Code Bench 20.1% #88 of 120 Official leaderboard
PaperBench 24% #11 of 15 Lab model card
TasteVal 11.6× #15 of 20 Official leaderboard
OpenAI-Proof Q&A 2.00% #7 of 9 Lab model card
DesignArena 3D Design 1,038 #114 of 136 Official leaderboard
FrontierCS 23.1 #12 of 14 Official leaderboard
MLE-bench 30 8.00% #7 of 8 Lab model card
MirrorCode 4.44% #39 of 39 Epoch AI run
SWE-bench-Live Verified 30.4% #2 of 2 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
Step Race 5 #1 of 59 Official leaderboard
Elimination Game 6 #2 of 56 Official leaderboard
SnakeBench 35 #9 of 188 Official leaderboard
SnorkelWordle 94% #1 of 19 Official leaderboard
MobileWorld 55.3% #1 of 18 Official leaderboard
Terminal-Bench 49.6% #1 of 18 tbench.ai, via Epoch AI
MCP-Universe 44.2% #2 of 30 Official leaderboard
MCP-Bench 0.75 #1 of 15 Official leaderboard
Meta Agents Research Environments (ARE) / Gaia2 42.1% #1 of 13 Official leaderboard
lmgame-Bench 75.1 #2 of 23 Official leaderboard
MM-BrowseComp 24.3% #1 of 11 Official leaderboard
AdaPlanBench 67.8% #1 of 10 Official leaderboard
MCPMark 52.6% #3 of 28 Official leaderboard
Agent Poker Bench 1,103 #2 of 17 Official leaderboard
OdysseyArena Energy Dispatch 23.3% #1 of 8 Official leaderboard
OdysseyArena Repo System 62.5% #1 of 8 Official leaderboard
OdysseyArena Turn on Lights 28.3% #1 of 8 Official leaderboard
LLM-WikiRace 15% #3 of 21 Official leaderboard
BrowseComp-Plus 70.1% #1 of 7 Official leaderboard
LiveResearchBench 73.7 #1 of 7 Official leaderboard
CompileBench 93.3% #3 of 20 Official leaderboard
GAIA 78.4% #10 of 64 Official leaderboard
LLM Chess 1,086 #22 of 123 Official leaderboard
VitaBench 2.0 44.1% #4 of 17 Official leaderboard
OdysseyArena AI Trading 17.3% #2 of 8 Official leaderboard
WARC-Bench 51.3% #2 of 8 Official leaderboard
ScienceBoard 24.2% #4 of 15 Official leaderboard
WebTailBench 45.1% #2 of 7 Official leaderboard
ITBench 51.9% #4 of 13 Official leaderboard
LisanBench 2247.0 #41 of 133 Official leaderboard
DeepResearch Bench 49.6% #7 of 22 evals.futuresearch.ai, via Epoch AI
Deep Research Bench (FutureSearch) 0.50 #7 of 21 Official leaderboard
AgentBench FC 52.2% #4 of 12 Official leaderboard
OSWorld-Verified 69.9% #12 of 33 Official leaderboard
tau2-bench Verified 79.9% #4 of 11 Official leaderboard
Terminal-Bench 2.0 49.6% #17 of 43 Official leaderboard
DeepPlanning 30.5% #6 of 15 Official leaderboard
DeepSearchQA 59.4% #4 of 10 Official leaderboard
Pencil Puzzle Bench 3.30% #24 of 53 Official leaderboard
tau2-bench 80.1% #36 of 79 Lab model card
Toolathlon 37.7% #24 of 50 Official leaderboard
Chess Puzzles 36% #46 of 93 Epoch AI run
Mystery Game Puzzles 23% #42 of 84 Epoch AI run
TextQuests 17.2 #14 of 28 Official leaderboard
Wiki Live Challenge 53.6 #1 of 2 Official leaderboard
EnterpriseOps-Gym 30.9% #7 of 12 Official leaderboard
ScreenSpot-Pro 18.5% #7 of 11 Official leaderboard
LMArena Search Arena 1,189 #25 of 37 Official leaderboard
C3-Bench 5.86% #21 of 30 Official leaderboard
BrowseComp 54.9% #60 of 81 Lab model card
X Browse (xAI) 24.2% #3 of 4 Official leaderboard
EBR-bench 12.7% #24 of 31 Epoch AI run

Professional work

BenchmarkScoreRankSource
HealthBench 67.2% #1 of 31 Lab model card
LEXam 70.2% #1 of 30 Official leaderboard
HealthBench Hard 46.2% #1 of 21 Lab model card
MedHELM 70.3 #1 of 13 Official leaderboard
FinQA (Snorkel OpenEnv environment) 81% #2 of 22 Official leaderboard
SnorkelFinance 81% #2 of 22 Official leaderboard
SnorkelUnderwrite 83.3% #3 of 27 Official leaderboard
BizFinBench v2 61.5% #1 of 9 Official leaderboard
CaseLaw v2 66.5% #6 of 53 Official leaderboard
LegalBench 86% #18 of 138 Official leaderboard
Finance Reasoning (Snorkel) 51% #4 of 21 Official leaderboard
Vals Finance Agent v1 46.9% #9 of 47 Official leaderboard
MedCode 49.6% #21 of 96 Official leaderboard
FailureSensorIQ 67.2% #6 of 24 Official leaderboard
TaxCalcBench 41.7% #5 of 17 Official leaderboard
PRBench 49% #16 of 48 Official leaderboard
Vending-Bench $3,579 #3 of 9 Official leaderboard
HealthBench Consensus 95.7% #6 of 17 Lab model card
TaxEval v2 73.4% #47 of 129 Official leaderboard
GDPval 38.8% #2 of 5 Official leaderboard
ProfBench 49.4% #30 of 72 Official leaderboard
MedScribe 83.7% #42 of 97 Official leaderboard
MortgageTax 65.5% #39 of 87 Official leaderboard
HealthBench Professional 46.2% #3 of 6 Official leaderboard
HealthBench Consensus Global Health 0% #1 of 2 Lab model card
HealthBench Consensus Urgent 0.40% #1 of 2 Lab model card
CorpFin v2 61.1% #60 of 119 Official leaderboard
Vals Finance Agent 52.1% #25 of 49 Official leaderboard
APEX-Agents 18.3% #5 of 8 Official leaderboard
StockBench 9.00 #9 of 13 Official leaderboard
Vals Index v1 49% #25 of 34 Official leaderboard
Remote Labor Index 1.67% #16 of 19 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
MMStar 75.7% #1 of 173 Official leaderboard
MMMU 84.2% #2 of 136 Official leaderboard
MM-Vet 77.6% #7 of 164 Official leaderboard
MMBench-EN v1.1 86.8% #9 of 178 Official leaderboard
MicroVQA 59.4% #1 of 19 Official leaderboard
SpaCE-10 53.4% #2 of 36 Official leaderboard
MM-HELIX 84.5% #1 of 17 Official leaderboard
LEGO-Puzzles 72% #1 of 16 Official leaderboard
MMSI-Bench 41.9% #2 of 28 Official leaderboard
MME-CC 40.3% #1 of 12 Official leaderboard
SpatialWorld Physical 14.4% #1 of 9 Official leaderboard
MathCanvas-Bench 43.5% #2 of 16 Official leaderboard
Blueprint-Bench 0.43 #1 of 8 Official leaderboard
ViVerBench 74.4% #1 of 8 Official leaderboard
RoboRewardBench 0.81 #3 of 22 Official leaderboard
VPCT 66% #4 of 27 VPCT leaderboard, via Epoch AI
MADQA 77.7% #3 of 18 Official leaderboard
MATH-Vision (MathVision) 72% #23 of 105 Official leaderboard
VISTA 49.7% #12 of 54 Official leaderboard
Progress-Bench 21.3 #2 of 9 Official leaderboard
VideoMME v2 (with subtitles) 37% #7 of 31 Official leaderboard
MMMU-Pro 78.4% #24 of 90 Lab model card
Math-VR 58.1% #7 of 26 Official leaderboard
GeoBench 81% #13 of 48 Official leaderboard
Video-MMMU 84.6% #10 of 36 Lab model card
GSM8K-V 30.1% #7 of 24 Official leaderboard
ERQA 65.7% #11 of 36 Lab model card
MeasureBench Real-world 19.8% #6 of 19 Official leaderboard
ChartMuseum 62.9% #7 of 21 Official leaderboard
SpatialWorld Digital 36.4% #3 of 9 Official leaderboard
MeasureBench Synthetic 16.9% #7 of 19 Official leaderboard
CharXiv 81.1% #29 of 75 Lab model card
Butter-Bench 30% #2 of 4 Official leaderboard
RoboSpatial-Home 52.5% #2 of 4 Official leaderboard
Vision Arena 1,211 #73 of 136 Official leaderboard
Math Kangaroo 2025 (MathArena) 78.8% #12 of 22 Official leaderboard
ZeroBench 7.00% #43 of 76 Official leaderboard
ClockBench 11.1% #26 of 37 Official leaderboard
OCR Arena 1,406 #8 of 11 Official leaderboard
vCX Benchmark 82.3% #33 of 42 Official leaderboard
Drone-Bench 37.8% #15 of 19 Official leaderboard
VoxelBench Image 1,168 #11 of 13 Official leaderboard
CompassArena Multimodal 1,069 #7 of 8 Official leaderboard
VoxelBench 1,313 #50 of 56 Official leaderboard
OCRBench v2 55.5% #13 of 14 Official leaderboard

Safety and security

BenchmarkScoreRankSource
HELM Safety 98.1 #5 of 75 Official leaderboard
Spiral-Bench v1.0 87.0 #1 of 13 Official leaderboard
AIR-Bench 2024 87.7% #7 of 82 Official leaderboard
BaxBench 54.3% #3 of 34 Official leaderboard
AutoBaxBench 45% #1 of 11 Official leaderboard
PCT (Polarized Contrastive Pairs) 35.3% #7 of 46 Official leaderboard
AthenaBench 66.1% #1 of 6 Official leaderboard
SUPERChem 39.6% #1 of 6 Official leaderboard
World-Class Bio 54.6% #19 of 74 Official leaderboard
Lakera B3 (Backbone Breaker Benchmark) 72% #10 of 35 Official leaderboard
Spiral-Bench v1.2 57.4 #6 of 21 Official leaderboard
Human Pathogen Capabilities Test 59.2% #26 of 86 Official leaderboard
FORTRESS 17% #23 of 73 Official leaderboard
SimpleSafetyTests 99.8% #25 of 75 Official leaderboard
Molecular Biology Capabilities Test 50.8% #38 of 90 Official leaderboard
Capture the Flag (Collegiate) 31% #6 of 13 Lab model card
Anthropic Red Team 99.1% #36 of 75 Official leaderboard
CyberGym 60.2% #17 of 34 Official leaderboard
BrowseSafe-Bench 85.5% #2 of 4 Official leaderboard
CharXiv Missing Image 9.00% #1 of 2 Lab model card
Capture the Flag (Professional) 27% #12 of 21 Lab model card
SCONE-bench $393,900,000 #4 of 7 Official leaderboard
nocot-bench (NCRI) 67.0 #112 of 185 Official leaderboard
Apollo Research strategic deception evaluations 3.97% #2 of 3 Lab model card
Challenging prompts from red teamers with biosafety-relevant PhDs 92.1% #2 of 3 Lab model card
ProtocolQA Open-Ended 30% #19 of 27 Lab model card
Anthropic political even-handedness eval 89% #4 of 5 Official leaderboard
Coding Deception 17% #5 of 6 Lab model card
Virology Capabilities Test (VCT) 41.4% #26 of 31 Lab model card
Tacit Knowledge and Troubleshooting 66% #26 of 30 Lab model card
Multimodal Troubleshooting Virology 42% #25 of 27 Lab model card
TroubleshootingBench 22% #19 of 20 Lab model card
BinaryAudit 1.75% #20 of 20 Official leaderboard
Gray Swan Agent Red Teaming (ART) benchmark 56.8% #14 of 14 Lab model card
Cyber Range 0% #10 of 10 Lab model card
Broken Search Tool 11% #9 of 9 Lab model card
Coding prompt injections 97% #4 of 4 Lab model card
Filtered, adversarial sample of production prompts 95.7% #3 of 3 Lab model card
Deception in production traffic 2.10% #2 of 2 Lab model card
Browsing prompt injections 99% #1 of 1 Lab model card
Expert Red Teaming for Violent Attack Planning 65.1% #1 of 1 Lab model card
Pairwise Bioweaponization Campaign 60% #1 of 1 Lab model card
Sycophancy evaluation (offline) 0.04 #1 of 1 Lab model card
Tool-calling prompt injections 99% #1 of 1 Lab model card
Virology Capabilities Test (text-only) 46% #1 of 1 Lab model card

Other

BenchmarkScoreRankSource
OpenVLM Leaderboard 79.9 #2 of 173 Official leaderboard
HELM Capabilities 80.7 #4 of 60 Official leaderboard
TutorBench 55.3% #4 of 20 Official leaderboard
ACCoRD 73.1 #36 of 152 Official leaderboard
Conceptual Reasoning Index 55.5 #52 of 151 Official leaderboard
CAIS Risk Index 44.9 #24 of 67 Official leaderboard
LMCA 40% #57 of 147 Official leaderboard
CAIS Text Capabilities Index 27.7 #38 of 68 Official leaderboard
CAIS Vision Capabilities Index 47.8 #30 of 53 Official leaderboard
SAGE (Vals) 43.7% #47 of 81 Official leaderboard
AGI Score (A Definition of AGI) 57% #1 of 1 Official leaderboard

Its lab's numbers put it at 147.5, 2.5 points below its ECI

The Self-Reported Index fits one score to the 23 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 145.0 to 150.0. How we compute it.

Its lab's numbers next to 14 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
HMMT 2025 100% HMMT Harvard-MIT mathematics tournament, GPT-5 pro (python) 88.3% Leaderboard, high +12 pts
World-Class Bio 47.6% SecureBio World Class Biology, mean of n=10 (Table 15) 54.6% Leaderboard, high −7.0 pts
MultiChallenge 69.6% Scale MultiChallenge** Multi-turn instruction following 63.2% Leaderboard, thinking +6.4 pts
Human Pathogen Capabilities Test 53.7% SecureBio Human Pathogen Capabilities Test, mean of n=10 (Table 15) 59.2% Leaderboard, high −5.5 pts
AIME 2025 100% AIME 2025 Competition math, GPT-5 pro (python) 95% Leaderboard, high +5.0 pts
GPQA Diamond 89.4% GPQA Diamond, GPT-5 pro (python) 86.2% Epoch AI run, high +3.2 pts
Molecular Biology Capabilities Test 48% SecureBio Molecular Biology Capabilities Test, mean of n=10 (Table 15) 50.8% Leaderboard, high −2.8 pts
SWE-bench Verified 74.9% SWE-bench Verified (n=477), GPT-5 with thinking 74.4% Leaderboard, Prometheus-v1.2.1 +0.5 pts
FrontierMath 32.1% FrontierMath, Tier 1-3, GPT-5 pro (python) 32.4% Leaderboard, high; task version 1.1.0, Epoch −0.3 pts
tau2-bench 80.1% Mean of Tau2-bench Function calling (airline) / Tau2-bench Function calling (retail) / Tau2-bench Function calling (telecom) 80% Leaderboard Same
Aider Polyglot 88% Aider Polyglot Multi-language code editing 88% Leaderboard, high, aider, diff edits Same
MMMU 84.2% MMMU College-level visual problem-solving 84.2% Leaderboard, thinking Same
MMMU-Pro 78.4% MMMU Pro*** Graduate-level visual problem-solving 78.4% Leaderboard, thinking Same
Video-MMMU 84.6% VideoMMMU Video-based multimodal reasoning (max frame 256) 84.6% Leaderboard Same

OpenAI's models over time

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

50 100 150 2023 2024 2025 2026 2027 GPT-6.1 Terra GPT-6 Luna GPT-5.4 Mini GPT-5.2 Pro GPT-5 GPT-5 nano GPT-4.1 nano 50 100 150 2023 2024 2025 2026 2027 GPT-6.1 Terra GPT-6 Luna

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 Aug 7, 2025

Its system cards

Every result in the longer documents the lab published with it. 50 of them aren't in the launch post.

GPT-5 System Card Aug 7, 2025

Capabilities

Health

Hallucination

Safety

Robustness

Bias

Alignment

Bio and chem risk

Cyber risk

AI R&D risk

Autonomy