o1 ranks first on 12 of the 113 benchmarks it has scores on.
Closed weights Flagship
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLU-Pro | 89.3% | #3 of 212 | Official leaderboard | |
| MMLU | 91.8% | #2 of 127 | Model card | |
| SEAL Multilingual | 1,134 | #1 of 19 | Official leaderboard | |
| SEAL Precise Instruction Following | 92.0 | #1 of 18 | Official leaderboard | |
| ECLeKTic | 38.7% | #3 of 35 | Official leaderboard | |
| LLM Confabulation Leaderboard | 11.7% | #6 of 58 | Official leaderboard | |
| Fiction.LiveBench | 83.3% | #8 of 57 | Epoch AI | |
| SuperGPQA | 60.2% | #10 of 58 | Official leaderboard | |
| SimpleQA | 47% | #14 of 59 | Model card | |
| Arena-Hard v2.0 | 61% | #6 of 23 | Official leaderboard | |
| MGSM | 89.3% | #14 of 52 | Model card | |
| BuzzBench | 59.3 | #7 of 23 | Official leaderboard | |
| LMArena (Elo) | 1,402 | #125 of 333 | Official leaderboard | |
| FACTS Grounding | 78.8% | #15 of 34 | Official leaderboard | |
| PersonQA | 55% | #5 of 9 | Model card | |
| SimpleQA Verified | 41.1% | #47 of 84 | Epoch AI | |
| SimpleQA Hallucination Rate | 44% | #7 of 12 | Model card | |
| Humanity's Last Exam | 7.96% | #35 of 47 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 2.40% | #118 of 148 | Official leaderboard | |
| Lech Mazur Creative Writing | 7.02 | #36 of 44 | Epoch AI |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| U-MATH | 86.8% | #1 of 30 | Official leaderboard | |
| ZebraLogic | 81% | #2 of 46 | Official leaderboard | |
| MATH Level 5 | 94.7% | #13 of 94 | Epoch AI | |
| PolyMath | 38.9% | #12 of 38 | Official leaderboard | |
| QuantBench | 60% | #1 of 3 | Model card | |
| MATH-500 | 96.4% | #27 of 67 | Model card | |
| GPQA Diamond | 76.8% | #95 of 219 | Epoch AI | |
| OTIS Mock AIME 2024-2025 | 73.3% | #93 of 202 | Epoch AI | |
| AIME 2024 | 79.2% | #44 of 87 | Model card | |
| FrontierMath | 9.31% | #38 of 70 | Epoch AI | |
| DTBench | 57.8% | #106 of 178 | Official leaderboard | |
| AIME 2025 | 81.7% | #33 of 53 | Official leaderboard | |
| SimpleBench | 40.1% | #68 of 100 | Official leaderboard | |
| PHYBench | 18% | #20 of 28 | Official leaderboard | |
| HMMT 2025 | 48.3% | #39 of 53 | Official leaderboard | |
| ARC-AGI | 30.7% | #76 of 96 | Epoch AI | |
| FrontierMath Tiers 1-3 v2 | 14.7% | #82 of 89 | Epoch AI |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BigCodeBench-Hard | 35.5% | #2 of 96 | Official leaderboard | |
| HumanEval-V | 40.6% | #1 of 24 | Official leaderboard | |
| Aider-Edit | 84.2% | #2 of 37 | Official leaderboard | |
| OpenAI Research Engineer Interview: Coding | 100% | #1 of 7 | Model card | |
| Aider Polyglot | 61.7% | #14 of 54 | Epoch AI | |
| SEAL Coding | 1,083 | #8 of 26 | Official leaderboard | |
| Multi-SWE-bench | 6.43% | #4 of 11 | Official leaderboard | |
| METR Time Horizon | 42 min | #12 of 31 | Official leaderboard | |
| LiveSQLBench | 25.2% | #17 of 36 | Official leaderboard | |
| WeirdML | 46.1% | #57 of 112 | Official leaderboard | |
| OpenAI Research Engineer Interview: Multiple Choice | 78% | #5 of 9 | Model card | |
| METR Time Horizon 1.1 | 39 min | #15 of 26 | Official leaderboard | |
| MLE-bench | 14% | #12 of 18 | Model card | |
| SWE-bench Verified | 48.9% | #129 of 150 | Model card | |
| TasteVal | 6.71× | #18 of 20 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ToolComp | 70.1% | #1 of 32 | Official leaderboard | |
| Step Race | 4 | #5 of 59 | Official leaderboard | |
| CRMArena | 64.3% | #1 of 9 | Official leaderboard | |
| Berkeley Function Calling Leaderboard (BFCL) | 67.8% | #10 of 67 | Official leaderboard | |
| tau-bench | 63.9% | #8 of 23 | Model card | |
| LLM Chess | 280 | #49 of 124 | Official leaderboard | |
| Galileo Agent Leaderboard | 0.88 | #15 of 31 | Official leaderboard | |
| Elimination Game | 4 | #31 of 56 | Official leaderboard | |
| GAIA | 49.8% | #32 of 57 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #31 of 53 | Official leaderboard | |
| Chess Puzzles | 10% | #74 of 83 | Epoch AI |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedXpertQA-MM | 56.3% | #1 of 9 | Official leaderboard | |
| CadEval | 56% | #4 of 15 | Epoch AI | |
| TaxEval v2 | 74.3% | #38 of 129 | Official leaderboard | |
| LegalBench | 80.4% | #83 of 138 | Official leaderboard | |
| Vals Finance Agent v1 | 20.8% | #33 of 47 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMVU | 75.5% | #1 of 28 | Official leaderboard | |
| MMMU | 78.2% | #11 of 136 | Official leaderboard | |
| CharXiv | 55.1% | #8 of 77 | Official leaderboard | |
| VisualPuzzles | 51.8% | #4 of 29 | Official leaderboard | |
| VHELM | 71.6 | #7 of 41 | Official leaderboard | |
| SpatialViz-Bench | 41.4% | #2 of 8 | Epoch AI | |
| GeoBench | 80% | #15 of 48 | Official leaderboard | |
| MATH-Vision (MathVision) | 60.3% | #35 of 105 | Official leaderboard | |
| VISTA | 45.3% | #27 of 54 | Official leaderboard | |
| EnigmaEval | 5.65% | #21 of 41 | Epoch AI | |
| MathVista | 71% | #39 of 64 | Model card | |
| Vision Arena | 1,193 | #83 of 136 | Official leaderboard | |
| VPCT | 37% | #22 of 28 | Epoch AI | |
| Drone-Bench | 31.9% | #17 of 19 | Official leaderboard | |
| ZeroBench | 0% | #64 of 66 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Multimodal Troubleshooting Virology | 59% | #1 of 26 | Model card | |
| BaxBench | 48.2% | #4 of 34 | Official leaderboard | |
| HELM Safety | 97.6 | #9 of 75 | Official leaderboard | |
| Standard Refusal Evaluation | 100% | #1 of 8 | Model card | |
| ChangeMyView | 89.1% | #1 of 5 | Model card | |
| AIR-Bench 2024 | 80% | #23 of 82 | Official leaderboard | |
| Challenging Refusal Evaluation | 92% | #3 of 9 | Model card | |
| FORTRESS | 19.4% | #25 of 66 | Official leaderboard | |
| MakeMeSay | 42% | #4 of 9 | Model card | |
| SimpleSafetyTests | 99% | #34 of 75 | Official leaderboard | |
| MakeMePay | 27% | #3 of 6 | Model card | |
| Contextual Nuclear Knowledge | 74% | #2 of 4 | Model card | |
| Political Persuasion Parallel Generation | 47.1% | #2 of 4 | Model card | |
| Radiological and Nuclear Expert Knowledge | 70% | #2 of 4 | Model card | |
| WildChat (toxic conversations) | 97.9% | #2 of 4 | Model card | |
| CoT Deception Monitoring | 0.17% | #1 of 2 | Model card | |
| Red Teaming Network pairwise safety comparison | 59.8% | #1 of 2 | Model card | |
| Tacit Knowledge Brainstorm (Open-Ended) | 51% | #1 of 2 | Model card | |
| Virology Capabilities Test (VCT) | 35.4% | #4 of 7 | Official leaderboard | |
| Jailbreak Augmented Examples | 100% | #3 of 5 | Model card | |
| Anthropic Red Team | 98.3% | #50 of 75 | Official leaderboard | |
| BioLP-bench | 33% | #6 of 9 | Model card | |
| Capture the Flag (High School) | 46% | #6 of 9 | Model card | |
| XSTest | 93.6% | #11 of 16 | Model card | |
| StrongREJECT | 72% | #16 of 22 | Model card | |
| Capture the Flag (Professional) | 13% | #16 of 21 | Model card | |
| Capture the Flag (Collegiate) | 13% | #10 of 13 | Model card | |
| Tacit Knowledge and Troubleshooting | 66% | #24 of 29 | Model card | |
| ProtocolQA Open-Ended | 24% | #22 of 25 | Model card | |
| Human Sourced Jailbreaks | 94% | #8 of 8 | Model card | |
| Production Jailbreaks | 99% | #5 of 5 | Model card | |
| Image input safety evaluations | 96.1% | #2 of 2 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 75.7% | #5 of 48 | Epoch AI | |
| ACCoRD | 64.2 | #75 of 150 | Official leaderboard | |
| Conceptual Reasoning Index | 37.8 | #101 of 149 | Official leaderboard | |
| LMCA | 22.3% | #101 of 139 | Epoch AI | |
| CAIS Vision Capabilities Index | 40.9 | #43 of 53 | Official leaderboard | |
| CAIS Risk Index | 64.0 | #54 of 65 | Official leaderboard | |
| CAIS Text Capabilities Index | 8.00 | #63 of 66 | Official leaderboard |
OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 45 of them aren't in the launch post.