- Agents' Last Exam (Sol, xhigh, code mode) 53.6
- Artificial Analysis Intelligence Index v4.1 (Sol, max) 58.9
- Artificial Analysis Coding Agent Index v1.1 (Sol, max) 80
- Terminal-Bench 2.1 (Sol Ultra, 4 agents) 91.9
- DeepSWE v1.1 (Sol, max) 72.7
- BrowseComp (Sol Ultra, 4 agents) 92.2
- GDPval-AA v2 (Sol, max) 1748
- OSWorld 2.0 (Sol, max) 62.6
- AutomationBench (Sol, max) 18.1
- GeneBench Pro (Sol, max) 28.7
- LifeSciBench (Sol, max) 59.9
- Management Consulting Tasks (Internal) 43.2
- SWE-Bench Pro 64.6
- MedChemBench (Internal) 48.3
- HealthBench Professional 60.5
- BenchCAD 70.6
- MMMU Pro with tools 84.6% (no tools 83.0%) 84.6
- GPQA Diamond 94.6
- FrontierMath Tier 1-3 (v2) 89
- FrontierMath Tier 4 (v2) 83
- Toolathlon 58
- OpenAI MRCR v2 8-needle 256K-512K 91.5% (512K-1M 73.8%) 91.5
- GraphWalks BFS 256K F1 90.7% (1M F1 77.1%) 90.7
- ARC-AGI-3 7.78
GPT-5.6 Sol
GPT-5.6 Sol ranks first on 34 of the 412 benchmarks with five or more models scored.
435
Benchmarks with a score
34
Ranked first, out of five or more models
104
In the top three
OpenAI Jul 2026
GPT-5.6 Sol
Closed weights Flagship
163.0
#13 of 256
Epoch Capabilities Index
Self-Reported Index 162.6 #19 of 404
- MMMU-Pro 84.6% #1 /90
- DeepSWE 72.7% #4 /28
- AutomationBench 28.8% #11 /42
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 | |
|---|---|---|---|---|
| AA-LCR Extended | 79.5% | #1 of 49 | Official leaderboard | |
| HLE Extended | 77.3% | #1 of 31 | Official leaderboard | |
| SEA-HELM | 77.0 | #3 of 71 | Official leaderboard | |
| OpenAI MRCR v2 | 83.5% | #2 of 47 | Official leaderboard | |
| LMArena (Elo) | 1,485 | #19 of 335 | Official leaderboard | |
| Creative Writing v3 | 80.8 | #8 of 120 | Official leaderboard | |
| FACTS Parametric | 80.7% | #3 of 45 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 358333.0 | #6 of 77 | Official leaderboard | |
| DesignArena ASCII Art | 1,305 | #7 of 77 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 81.7 | #11 of 119 | Official leaderboard | |
| FACTS Search V2 | 88.5% | #2 of 21 | Official leaderboard | |
| ELI5 Arena | 1,689 | #2 of 18 | Official leaderboard | |
| Extended NYT Connections | 93.8% | #12 of 103 | Official leaderboard | |
| Document Arena | 1,486 | #5 of 39 | Official leaderboard | |
| FACTS Parametric V2 | 81.1% | #4 of 27 | Official leaderboard | |
| Lech Mazur Creative Writing | 2.42 | #8 of 50 | Official leaderboard | |
| SimpleQA Verified | 69.2% | #9 of 56 | Official leaderboard | |
| StudyArena Text | 1,263 | #18 of 85 | Official leaderboard | |
| Hemingway-bench | 1,057 | #9 of 42 | Official leaderboard | |
| Last Translation Benchmark | 28.1 | #8 of 36 | Official leaderboard | |
| GraphWalks | 90.7% | #6 of 27 | Lab model card | |
| Global MMLU Lite | 94.1% | #11 of 43 | Official leaderboard | |
| Opening Lines Arena | 1,576 | #4 of 15 | Official leaderboard | |
| Joke Arena | 1,519 | #8 of 25 | Official leaderboard | |
| FACTS Benchmark Suite | 70.6% | #8 of 23 | Official leaderboard | |
| Continual Learning Exam | 33.2% | #4 of 11 | Official leaderboard | |
| EQ-Bench 4 | 1,236 | #13 of 32 | Official leaderboard | |
| BullshitBench V2 | 48% | #56 of 129 | Official leaderboard | |
| Antidote | 1,027 | #18 of 36 | Official leaderboard | |
| HLE-Diamond | 41% | #10 of 20 | Official leaderboard | |
| BullshitBench | 27.3% | #55 of 107 | Official leaderboard | |
| FACTS Multimodal | 46.4% | #13 of 23 | Official leaderboard | |
| Corp Speak Arena | 1,481 | #13 of 22 | Official leaderboard | |
| HUMAINE | 23 | #35 of 50 | Official leaderboard | |
| FACTS Grounding v2 | 61.9% | #18 of 25 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 12.4% | #85 of 104 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 100% | #4 of 212 | Epoch AI run | |
| Riemann-bench | 83.2% | #1 of 20 | Official leaderboard | |
| ProofBench | 77% | #3 of 59 | Official leaderboard | |
| ARC-AGI-2 | 92.5% | #6 of 98 | Official leaderboard | |
| ForecastBench | 0.12 | #6 of 82 | Official leaderboard | |
| GPQA Diamond | 93.5% | #19 of 229 | Epoch AI run | |
| Verifying the Verifiers (Project APE CRED benchmark) | 99% | #2 of 23 | Official leaderboard | |
| ASI-Bench | 55.8 | #1 of 11 | Official leaderboard | |
| FrontierChallenge | 20.6% | #1 of 11 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 89.1% | #9 of 86 | Official leaderboard | |
| ARC-AGI | 97.5% | #10 of 95 | Official leaderboard | |
| DTBench | 95.7% | #20 of 180 | Official leaderboard | |
| Surface Evolver Bench | 93.1% | #3 of 25 | Official leaderboard | |
| HorizonMath | 11.5% | #1 of 8 | Official leaderboard | |
| ResearchClawBench | 34.8 | #3 of 23 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 82.9% | #10 of 68 | Official leaderboard | |
| Leipzig Benchmark | 90% | #3 of 19 | Official leaderboard | |
| PHYSICS-Verified | 90.4% | #3 of 19 | Official leaderboard | |
| ExplorationBench | 88.1% | #2 of 10 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 59.8% | #7 of 26 | Official leaderboard | |
| PhysClosureBench | 5.83 | #2 of 7 | Official leaderboard | |
| ProofBench v1.1 | 83% | #15 of 52 | Official leaderboard | |
| SimpleBench | 64.8% | #31 of 106 | Official leaderboard | |
| Terminal-Bench-Science | 22.4% | #5 of 17 | Official leaderboard | |
| BixBench3 | 48.0 | #6 of 20 | Official leaderboard | |
| Into the Unknown | 0% | #3 of 9 | Official leaderboard | |
| CritPt | 74.9% | #4 of 11 | Official leaderboard | |
| CritPt v2 | 74.9% | #4 of 11 | Official leaderboard | |
| Human Behavior Bench | 79.5 | #4 of 11 | Official leaderboard | |
| PaperBenchX | 9.68% | #4 of 10 | Official leaderboard | |
| FrontierPhysics | 7.10% | #9 of 22 | Official leaderboard | |
| MysteryMechanism | 33.3% | #13 of 31 | Official leaderboard | |
| BioMysteryBench | 71.1% | #13 of 30 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 43.3% | #13 of 27 | Official leaderboard | |
| BTF-3 | 0.14 | #4 of 8 | Official leaderboard | |
| EurekaBench | 19.4 | #4 of 7 | Official leaderboard | |
| FrontierMath Erdős | 0% | #10 of 15 | Official leaderboard | |
| Apodex Discovery | 2.27 | #7 of 8 | Official leaderboard | |
| Prophet Arena | 84.8 | #22 of 24 | Official leaderboard | |
| SciUniverse | 9.40% | #6 of 6 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IOI (Vals) | 86.7% | #2 of 55 | Official leaderboard | |
| SlopCodeBench | 32.1% | #1 of 22 | Official leaderboard | |
| WeirdML | 89.4% | #6 of 112 | Official leaderboard | |
| DesignArena Game Dev | 1,365 | #9 of 143 | Official leaderboard | |
| DesignArena Website | 1,320 | #10 of 151 | Official leaderboard | |
| DesignArena SVG | 1,305 | #7 of 103 | Official leaderboard | |
| FrontierCS | 76.4 | #1 of 14 | Official leaderboard | |
| DesignArena Code | 1,339 | #12 of 143 | Official leaderboard | |
| DAXBench v2 | 91.5% | #4 of 47 | Official leaderboard | |
| DesignArena Data Visualization | 1,322 | #11 of 118 | Official leaderboard | |
| AndroidBench | 90.8% | #3 of 32 | Official leaderboard | |
| DesignArena 3D Design | 1,372 | #13 of 136 | Official leaderboard | |
| ALE-Bench | 2,410 | #13 of 126 | Official leaderboard | |
| ReactBench | 46.7% | #3 of 28 | Official leaderboard | |
| IOI v2 (Vals) | 91.2% | #5 of 45 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,213 | #3 of 27 | Official leaderboard | |
| DesignArena UI Component | 1,327 | #17 of 139 | Official leaderboard | |
| GSO-Bench | 76.5% | #4 of 30 | Official leaderboard | |
| DeepSWE | 72.7% | #4 of 28 | Official leaderboard | |
| WebDev Arena | 1,617 | #16 of 105 | arena.ai, via Epoch AI | |
| Code Migration | 52.9% | #12 of 78 | Official leaderboard | |
| SWE-bench-Live | 70.3% | #2 of 12 | Official leaderboard | |
| Legacy-Bench | 60% | #1 of 6 | Official leaderboard | |
| Senior SWE-Bench | 34.7% | #4 of 23 | Official leaderboard | |
| DesignArena React Native | 1,228 | #11 of 56 | Official leaderboard | |
| SWE-sweep | 4.70% | #2 of 10 | Official leaderboard | |
| QwenSVGBench | 1,758 | #1 of 5 | Official leaderboard | |
| SWE-Bench Pro | 64.6% | #17 of 81 | Lab model card | |
| Doc2Harness | 43.8% | #4 of 19 | Official leaderboard | |
| DesignArena Full Stack | 1,237 | #11 of 49 | Official leaderboard | |
| Code Arena | 1,615 | #29 of 128 | Official leaderboard | |
| DesignArena Image to HTML | 1,231 | #12 of 51 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,219 | #11 of 46 | Official leaderboard | |
| SWE Refactor Bench | 28.5% | #2 of 8 | Official leaderboard | |
| Image-to-WebDev Arena | 1,606 | #14 of 55 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 24.7% | #10 of 38 | Official leaderboard | |
| DesignArena Agentic Overall | 1,233 | #20 of 74 | Official leaderboard | |
| PostTrainBench 1.1 | 36.2% | #3 of 11 | Official leaderboard | |
| Vibe Code Bench | 80.5% | #33 of 120 | Official leaderboard | |
| APEX-SWE | 48% | #18 of 65 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,239 | #7 of 24 | Official leaderboard | |
| GameDevBench | 63.7% | #5 of 17 | Official leaderboard | |
| SWE-bench Science | 40.3% | #3 of 10 | Official leaderboard | |
| SWE Atlas | 46% | #7 of 22 | Official leaderboard | |
| SWE Atlas Test Writing | 45.9% | #7 of 22 | Official leaderboard | |
| ProgramBench | 1.00% | #8 of 25 | Official leaderboard | |
| DesignArena Frontend Web App | 1,226 | #17 of 53 | Official leaderboard | |
| DesignArena Image to Frontend | 1,235 | #10 of 31 | Official leaderboard | |
| GBAEval | 52.6% | #7 of 21 | Official leaderboard | |
| RSI-Exam | 43.3% | #6 of 18 | Official leaderboard | |
| GameCraft-Bench | 60.5% | #5 of 15 | Official leaderboard | |
| MiMo Visual Coding | 73.4% | #3 of 9 | Official leaderboard | |
| Artificial Analysis Coding Agent Index v1.1 | 80.0 | #1 of 3 | Lab model card | |
| SWE-bench Verified Extended | 63.3% | #15 of 44 | Official leaderboard | |
| TasteVal | 59.0× | #7 of 20 | Official leaderboard | |
| ProgramBench almost solved | 15.5% | #9 of 25 | Official leaderboard | |
| ProgramBench tests passed | 69.9% | #9 of 25 | Official leaderboard | |
| FrontierCode 2.0 | 28.9% | #12 of 32 | Official leaderboard | |
| SWE-rebench | 62.3% | #5 of 13 | Official leaderboard | |
| MLS-Bench | 46.2% | #6 of 15 | Official leaderboard | |
| τ^τ-bench | 22% | #2 of 5 | Official leaderboard | |
| MacroscopeBench | 77.8% | #7 of 16 | Official leaderboard | |
| SWE-Clar | 57% | #4 of 9 | Official leaderboard | |
| Vibe Code Bench 1-100 | 20% | #12 of 26 | Official leaderboard | |
| FrontierSWE | 32.2% | #12 of 25 | Official leaderboard | |
| WeirdML v3 | 15% | #13 of 27 | Official leaderboard | |
| QwenBlenderBench | 68.6% | #3 of 6 | Official leaderboard | |
| ReviewBench | 70.2% | #2 of 4 | Official leaderboard | |
| OpenRSI Index Isaac Lab PegInsert Reward Search | 0.60 | #1 of 2 | Official leaderboard | |
| OpenRSI Index Kev Decision Architecture | 0.29 | #1 of 2 | Official leaderboard | |
| OpenRSI Index Marin Scaling Ladder | 1.01 | #1 of 2 | Official leaderboard | |
| OpenRSI Index MInference 32-Head Sparse Prefill Kernel | 1.54× | #1 of 2 | Official leaderboard | |
| Vals RSI Index | 0.24 | #14 of 27 | Official leaderboard | |
| AutoResearchExam | 0.51 | #7 of 12 | Official leaderboard | |
| CursorBench | 41.7% | #13 of 21 | cursor.com, via Epoch AI | |
| PostTrainBench 1.2 | 31% | #13 of 21 | Official leaderboard | |
| CursorBench 4.0 | 41.7% | #14 of 21 | Official leaderboard | |
| SWE-Bench Pro V2 | 95.5% | #12 of 18 | Official leaderboard | |
| SWE-Bench Pro V2 Hard | 82.4% | #12 of 18 | Official leaderboard | |
| TestPrism | 23.3% | #4 of 6 | Official leaderboard | |
| Brokk Power Ranking | 7.60 | #13 of 19 | Official leaderboard | |
| Vision2Web | 73.5% | #7 of 10 | Official leaderboard | |
| OpenDesign Arena | 77.6 | #15 of 21 | Official leaderboard | |
| QwenQoderBench | 53.8% | #4 of 5 | Official leaderboard | |
| QwenReactBench | 1,564 | #4 of 5 | Official leaderboard | |
| QwenSWEBench | 73.5% | #4 of 5 | Official leaderboard | |
| SWE-Together | 57.8% | #19 of 22 | Official leaderboard | |
| LibraryDesignBench | 39.5 | #7 of 8 | Official leaderboard | |
| MiMo Code Bench | 59.3% | #7 of 8 | Official leaderboard | |
| MirrorCode | 20% | #37 of 39 | Epoch AI run | |
| OpenRSI Index ACE Playbook Inspection and Repair | 0.70 | #2 of 2 | Official leaderboard | |
| OpenRSI Index Learnability-Aware Long/Short CoT Adaptation | 46.1% | #2 of 2 | Official leaderboard | |
| OpenRSI Index Molmo2 Video-Pointing Inference Strategy | 0.46 | #2 of 2 | Official leaderboard | |
| OpenRSI Index MolmoWeb Interaction Context Allocation | 0.61 | #2 of 2 | Official leaderboard | |
| OpenRSI Index ReasonIR Difficulty Curriculum | 0.22 | #2 of 2 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseComp | 92.2% | #2 of 81 | Lab model card | |
| LMArena Search Arena | 1,255 | #1 of 37 | Official leaderboard | |
| Game Arena Poker | 30.3 | #1 of 31 | Official leaderboard | |
| WildClawBench | 67.2% | #1 of 31 | Official leaderboard | |
| Terminal-Bench 2.1 | 91.9% | #2 of 56 | Lab model card | |
| LLM Chess | 1,550 | #5 of 123 | Official leaderboard | |
| Chess Puzzles | 68% | #4 of 93 | Epoch AI run | |
| Long-Horizon-Terminal-Bench (LHTB) | 60% | #2 of 25 | Official leaderboard | |
| CAR-bench | 70% | #1 of 12 | Official leaderboard | |
| BALROG | 60% | #3 of 33 | Official leaderboard | |
| Web Context (Nimble) | 80% | #1 of 11 | Official leaderboard | |
| BabaIsBench | 100% | #3 of 32 | Official leaderboard | |
| Bench on the Clocktower | 58% | #1 of 10 | Official leaderboard | |
| Toolathlon | 58% | #3 of 29 | Lab model card | |
| Mystery Game Puzzles | 58% | #10 of 84 | Epoch AI run | |
| Long-Horizon-Terminal-Bench 2.0 | 67.2% | #2 of 15 | Official leaderboard | |
| ScienceBoard | 42.6% | #2 of 15 | Official leaderboard | |
| GameHorizon | 74.8 | #5 of 37 | Official leaderboard | |
| CEO-Bench | $11,313,982 | #3 of 19 | Official leaderboard | |
| τ³-bench Banking | 46.9% | #4 of 25 | Official leaderboard | |
| Terminal-Bench 3.0 | 34.6% | #2 of 11 | Official leaderboard | |
| RuneBench | 5.90 | #14 of 76 | Official leaderboard | |
| LisanBench | 4699.0 | #25 of 133 | Official leaderboard | |
| DuelLab GameBench 2 | 80.1 | #11 of 57 | Official leaderboard | |
| Game Arena Dark Hex | 354 | #8 of 41 | Official leaderboard | |
| Game Arena Word Association | 350 | #8 of 40 | Official leaderboard | |
| PinchBench 2.0 | 84.2% | #12 of 57 | Official leaderboard | |
| CLASHarness | 24.5% | #3 of 14 | Official leaderboard | |
| Era by Eon | 69.7% | #2 of 9 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 443 | #10 of 39 | Official leaderboard | |
| AutomationBench | 28.8% | #11 of 42 | Official leaderboard | |
| ApprenticeBench | 26% | #5 of 18 | Official leaderboard | |
| Game Arena Lines of Action | 522 | #11 of 39 | Official leaderboard | |
| Game Arena Word Art | 1,106 | #11 of 39 | Official leaderboard | |
| Agent Arena | 4.63% | #16 of 56 | Official leaderboard | |
| RecreationBench | 42.1% | #3 of 10 | Official leaderboard | |
| Game Arena Bargaining | 218 | #12 of 39 | Official leaderboard | |
| Game Arena Coin Game | 416 | #12 of 39 | Official leaderboard | |
| Game Arena Chess | 1,132 | #13 of 42 | Official leaderboard | |
| Kaggle Game Arena | 356 | #14 of 42 | Official leaderboard | |
| TraceTrial-Bench | 76.1% | #3 of 9 | Official leaderboard | |
| Brood War Bench | 72.2% | #2 of 6 | Official leaderboard | |
| EBR-bench | 44.8% | #11 of 31 | Epoch AI run | |
| BrowseComp Extended | 32.3% | #15 of 42 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 37.4% | #16 of 44 | Official leaderboard | |
| Clocktower Bench | 1,591 | #4 of 11 | Official leaderboard | |
| Learn2Play Bench | 74.3 | #4 of 11 | Official leaderboard | |
| OSWorld-Science | 50.1% | #4 of 11 | Official leaderboard | |
| Terminal-Bench 4.1 | 38.5% | #16 of 43 | Official leaderboard | |
| MCPMark Verified | 92.9% | #3 of 8 | Official leaderboard | |
| Realtime-Bench | 18% | #3 of 8 | Official leaderboard | |
| MCP Atlas | 81.8% | #14 of 37 | Official leaderboard | |
| Time Horizon Index: KSP | 23.8% | #5 of 13 | Official leaderboard | |
| Vals Web Search Index | 49.1% | #2 of 5 | Official leaderboard | |
| ARC-AGI-3 | 7.78% | #11 of 27 | Official leaderboard | |
| Game Arena Clobber | 339 | #16 of 39 | Official leaderboard | |
| OSWorld 2.0 | 62.6% | #8 of 19 | Lab model card | |
| Terminal-Bench 4.0 | 37.3% | #17 of 40 | Official leaderboard | |
| MyPCBench | 55.4% | #4 of 9 | Official leaderboard | |
| Game Arena Dots and Boxes | 923 | #18 of 40 | Official leaderboard | |
| Game Arena Go | 1,005 | #18 of 39 | Official leaderboard | |
| Game Arena Reversi | 585 | #18 of 39 | Official leaderboard | |
| SnakeBench | 25 | #90 of 188 | Official leaderboard | |
| Game Arena Nine Men's Morris | 637 | #19 of 39 | Official leaderboard | |
| MobileWorld | 40.8% | #9 of 18 | Official leaderboard | |
| SaaS-Bench | 29.2% | #2 of 4 | Official leaderboard | |
| Game Arena Four in a Row | 353 | #23 of 43 | Official leaderboard | |
| CUA-bench | 8.33% | #5 of 9 | Official leaderboard | |
| Game Arena Checkers | 885 | #27 of 41 | Official leaderboard | |
| DuetBench | 990 | #7 of 9 | Official leaderboard | |
| TextQuests | 26.9 | #26 of 28 | Official leaderboard | |
| OSWorld 2.0 (Aug 2026) | 27.3% | #3 of 3 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| E-Commerce Bench | 14.3 | #1 of 19 | Official leaderboard | |
| ProfBench | 60.4% | #4 of 72 | Official leaderboard | |
| TRLawBench | 9.65 | #2 of 34 | Official leaderboard | |
| LegalBench | 87% | #9 of 138 | Official leaderboard | |
| PhysicianBench | 50% | #1 of 13 | Official leaderboard | |
| PolicyBench | 93.1% | #4 of 50 | Official leaderboard | |
| Excel Modeling Benchmark | 72.3% | #7 of 76 | Official leaderboard | |
| MMEnterprise-Bench | 68% | #1 of 9 | Official leaderboard | |
| Vals Index v2 | 63.7% | #8 of 65 | Official leaderboard | |
| E-Bench (Tencent) | 80.6% | #1 of 8 | Official leaderboard | |
| E-Bench-Code (Tencent) | 83.3% | #1 of 8 | Official leaderboard | |
| APEX-1 | 73.1% | #9 of 71 | Official leaderboard | |
| Legal Research Bench | 48.1% | #13 of 93 | Official leaderboard | |
| Management Consulting Tasks (OpenAI) | 43.2% | #1 of 7 | Official leaderboard | |
| MedXpertQA MM Extended | 64.7% | #6 of 41 | Official leaderboard | |
| Vending-Bench 2 | $9,619 | #11 of 67 | Official leaderboard | |
| FoodTruck Bench | $53,229 | #10 of 59 | Official leaderboard | |
| DesignArena HTML Slides | 1,181 | #5 of 28 | Official leaderboard | |
| OR-License | 84.5% | #2 of 11 | Official leaderboard | |
| TaxCalcBench v2 | 72% | #6 of 32 | Official leaderboard | |
| Agents' Last Exam | 30.6% | #6 of 31 | Official leaderboard | |
| RCM-Bench-1 | 91.2% | #4 of 20 | Official leaderboard | |
| BankerToolBench v1.1 | 29% | #2 of 10 | Official leaderboard | |
| PRBench | 50.5% | #10 of 48 | Official leaderboard | |
| CHI-Bench | 25.3% | #5 of 23 | Official leaderboard | |
| LMBuild | 64.9% | #4 of 18 | Official leaderboard | |
| BlenderWorld | 67.7 | #3 of 13 | Official leaderboard | |
| TaxEval v2 | 74.8% | #30 of 129 | Official leaderboard | |
| Vals Index v2.1 | 58% | #12 of 49 | Official leaderboard | |
| OfficeQA Pro | 60.2% | #3 of 12 | Official leaderboard | |
| MerchantBench | 52.9 | #2 of 8 | Official leaderboard | |
| GDPval-AA v2 | 1,748 | #7 of 27 | Lab model card | |
| Tuesday Work Index | 61.1 | #12 of 44 | Official leaderboard | |
| EngiWorld | 38.6% | #2 of 7 | Official leaderboard | |
| MortgageTax | 67.3% | #26 of 87 | Official leaderboard | |
| APEX-Accounting | 62.2% | #15 of 49 | Official leaderboard | |
| Vending-Bench Arena | $7,821 | #10 of 32 | Official leaderboard | |
| MedScribe | 85.2% | #31 of 97 | Official leaderboard | |
| CorpFin v2 | 64.4% | #39 of 119 | Official leaderboard | |
| Tax Agent Bench | 31.1% | #30 of 84 | Official leaderboard | |
| Public Benefits Bench | 66.5% | #18 of 50 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,198 | #13 of 36 | Official leaderboard | |
| CommerceAgentBench | 51.4% | #5 of 13 | Official leaderboard | |
| Vals Finance Agent v2 | 53.8% | #31 of 80 | Official leaderboard | |
| LifeSci Diamond | 5.50% | #7 of 18 | Official leaderboard | |
| GDPval Extended | 1.30% | #17 of 43 | Official leaderboard | |
| DesignArena Spreadsheet | 1,210 | #14 of 35 | Official leaderboard | |
| HW-Bench | 30% | #6 of 15 | Official leaderboard | |
| DAYJOB: Finance | 11.8% | #16 of 39 | Official leaderboard | |
| Bedside Bench | 91.1% | #7 of 17 | Official leaderboard | |
| MedCode | 44% | #41 of 96 | Official leaderboard | |
| JobBench | 45.4% | #7 of 16 | Official leaderboard | |
| Short Form Bench | $2,410 | #4 of 9 | Official leaderboard | |
| ATLAS-Finance | 4.00% | #5 of 11 | Official leaderboard | |
| ATLAS-Finance v1.1.0 | 4.00% | #5 of 11 | Official leaderboard | |
| DAYJOB: Healthcare | 4.40% | #17 of 37 | Official leaderboard | |
| FrontierFinance | 46.8% | #8 of 17 | Official leaderboard | |
| HealthBench Consensus | 95.5% | #8 of 17 | Lab model card | |
| HealthBench Hard | 33.1% | #11 of 21 | Lab model card | |
| EnterpriseBench: CoreCraft | 46.2% | #27 of 51 | Official leaderboard | |
| APEX-Agents | 51.4% | #33 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 65.2% | #36 of 65 | Official leaderboard | |
| OfficeQA Pro V2 | 33.3% | #4 of 7 | Official leaderboard | |
| HealthBench Professional | 60.5% | #15 of 26 | Lab model card | |
| Harvey LAB (Legal Agent Benchmark) | 2.50% | #47 of 80 | Official leaderboard | |
| DesignArena PDF | 1,173 | #22 of 37 | Official leaderboard | |
| Genspark Slides Benchmark | 67% | #6 of 10 | Official leaderboard | |
| BenchCAD | 70.6% | #7 of 11 | Lab model card | |
| Parametric CAD Bench | 30% | #7 of 11 | Official leaderboard | |
| YC-Bench | $728,001 | #41 of 59 | Official leaderboard | |
| HealthBench | 57% | #22 of 31 | Lab model card | |
| Agents' Last Exam v1.1 | 30.3% | #11 of 14 | Official leaderboard | |
| CoWorkBench | 71.5% | #4 of 5 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU-Pro | 84.6% | #1 of 90 | Lab model card | |
| ZeroBench | 54.6% | #1 of 76 | Official leaderboard | |
| PerceptionBench (Moonshot) | 59.7% | #1 of 16 | Official leaderboard | |
| ParseBench | 74.4% | #3 of 44 | Official leaderboard | |
| MMMU-Pro Extended | 51.9% | #3 of 38 | Official leaderboard | |
| Vision Arena | 1,288 | #14 of 136 | Official leaderboard | |
| BabyVision | 88.9% | #3 of 27 | Official leaderboard | |
| AgenticVBench | 38.4% | #1 of 9 | Official leaderboard | |
| VoxelBench | 2,252 | #7 of 56 | Official leaderboard | |
| RoTA-Bench | 17% | #1 of 8 | Official leaderboard | |
| ClockBench | 68.3% | #5 of 37 | Official leaderboard | |
| Chartography | 45% | #6 of 43 | Official leaderboard | |
| ExtractBench | 93.8% | #5 of 27 | Official leaderboard | |
| BVB | 71.8 | #9 of 45 | Official leaderboard | |
| Code2Video Benchmark | 1,548 | #4 of 19 | Official leaderboard | |
| CharXiv Extended | 82.9% | #9 of 38 | Official leaderboard | |
| ArtifactArena | 1,165 | #7 of 21 | Official leaderboard | |
| InteractiveBench | 30.8% | #4 of 12 | Official leaderboard | |
| Furniture Assembly Benchmark | 56.7% | #12 of 35 | Official leaderboard | |
| Drone-Bench | 76.1% | #7 of 19 | Official leaderboard | |
| GlobeBench | 23.6% | #7 of 19 | Official leaderboard | |
| 4DCodeBench | 0.65 | #6 of 16 | Official leaderboard | |
| Blueprint-Bench 2 | 33.6% | #15 of 38 | Official leaderboard | |
| EnigmaEval | 37.1% | #2 of 5 | Official leaderboard | |
| Instrument reading (Google DeepMind) | 61.5% | #2 of 5 | Official leaderboard | |
| Progress classification (Google DeepMind) | 46.2% | #2 of 5 | Official leaderboard | |
| Human proximity (Google DeepMind robotics) | 83.4% | #2 of 4 | Official leaderboard | |
| GeoBench | 73% | #25 of 48 | Official leaderboard | |
| DesignArena Code to Video | 1,137 | #16 of 29 | Official leaderboard | |
| CodeCap-Bench | 50.0 | #4 of 7 | Official leaderboard | |
| Humanity's Sixth Sense | 30% | #22 of 34 | Official leaderboard | |
| VibeVideoBench | 3.32 | #6 of 9 | Official leaderboard | |
| OmniParseBench | 87.8 | #7 of 9 | Official leaderboard | |
| 1H-VideoQA | 62.4% | #17 of 20 | Official leaderboard | |
| RoboDojo-RC Tier 1 | 14.9% | #8 of 9 | Official leaderboard | |
| CineRecon | 35.9 | #11 of 12 | Official leaderboard | |
| QwenVisualOffice | 29.5% | #6 of 6 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CyberBench (Vals) | 88.1% | #1 of 26 | Official leaderboard | |
| Molecular Biology Capabilities Test | 61.1% | #4 of 90 | Official leaderboard | |
| World-Class Bio | 64.4% | #4 of 74 | Official leaderboard | |
| Human Pathogen Capabilities Test | 76.3% | #5 of 86 | Official leaderboard | |
| nocot-bench (NCRI) | 109.4 | #13 of 185 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 70.6% | #5 of 71 | Official leaderboard | |
| KORA | 68.7% | #3 of 42 | Official leaderboard | |
| CVE-Bench | 94.1% | #1 of 14 | Lab model card | |
| GeneBench Pro | 31.5% | #1 of 12 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 76.3% | #4 of 44 | Official leaderboard | |
| ProtocolQA Open-Ended | 44.1% | #3 of 27 | Lab model card | |
| scBench-Long | 29.6% | #3 of 27 | Official leaderboard | |
| dfbench | 65.7% | #2 of 18 | Official leaderboard | |
| SRE-Bench | 30.5% | #5 of 35 | Official leaderboard | |
| Agent JSK | 100% | #1 of 7 | Lab model card | |
| PatchEval | 83.9% | #1 of 7 | Official leaderboard | |
| PatchEval-Verified | 83.9% | #1 of 7 | Official leaderboard | |
| Tacit Knowledge and Troubleshooting | 82.2% | #5 of 30 | Lab model card | |
| SpatialBench-Long | 37.5% | #5 of 27 | Official leaderboard | |
| VCT-v2 | 49.4% | #5 of 25 | Official leaderboard | |
| TroubleshootingBench | 48% | #4 of 20 | Lab model card | |
| Cyber Range | 83.3% | #2 of 10 | Lab model card | |
| xdev-bench-hard | 100% | #1 of 5 | Official leaderboard | |
| DeepsecBench | 35.3 | #6 of 29 | Official leaderboard | |
| ExploitGym | 33.7% | #4 of 18 | Official leaderboard | |
| Capture the Flag (Internal) | 97.1% | #1 of 4 | Lab model card | |
| Hard-negative protein binding prediction | 7.60% | #1 of 4 | Lab model card | |
| Multimodal Troubleshooting Virology | 55.5% | #7 of 27 | Lab model card | |
| ABC Bench Liquid-Handling Robot | 98.7% | #9 of 34 | Official leaderboard | |
| CyberGym | 83.6% | #9 of 34 | Lab model card | |
| TxBench-Preclinical-Pharmacology | 61.3% | #8 of 30 | Official leaderboard | |
| Dictatorship Eval | 85.4% | #6 of 20 | Official leaderboard | |
| DrugDiscoveryBench | 56.5% | #6 of 20 | Official leaderboard | |
| EpiBench | 25.8% | #9 of 27 | Official leaderboard | |
| WellnessBench | 57% | #4 of 12 | Official leaderboard | |
| Last Secure Code Benchmark | 24.7% | #2 of 6 | Official leaderboard | |
| DistressBench | 73.6% | #15 of 42 | Official leaderboard | |
| HackBench | 53.0 | #3 of 8 | Official leaderboard | |
| OncologyBench | 18.7% | #12 of 30 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 41.7% | #19 of 46 | Official leaderboard | |
| BioTIER Refuse | 74% | #29 of 69 | Official leaderboard | |
| FORTRESS | 20.1% | #31 of 73 | Official leaderboard | |
| AAV Capsid Packaging Prediction (OpenAI) | 0.53 | #3 of 7 | Lab model card | |
| BioSecBench-Surveillance | 47.3% | #12 of 27 | Official leaderboard | |
| CyScenarioBench | 28% | #8 of 17 | Lab model card | |
| VariantBench | 35.3% | #13 of 27 | Official leaderboard | |
| Virology Capabilities Test (VCT) | 53.5% | #15 of 31 | Lab model card | |
| TxBench-Oligo-Discovery | 37.2% | #15 of 30 | Official leaderboard | |
| SecIT Bench | 75.4% | #13 of 26 | Official leaderboard | |
| Destructive action avoidance | 83% | #4 of 8 | Lab model card | |
| UK AISI cyber ranges | 70% | #2 of 4 | Lab model card | |
| ReproBAIT | 85% | #1 of 2 | Lab model card | |
| UK AISI narrow cyber tasks | 95% | #1 of 2 | Lab model card | |
| BioTIER Permit | 99.3% | #38 of 69 | Official leaderboard | |
| BioSecBench-Function | 35.5% | #15 of 27 | Official leaderboard | |
| TxBench-Antibody-Discovery | 37.3% | #17 of 30 | Official leaderboard | |
| VIALS 2 | 26.5% | #8 of 14 | Official leaderboard | |
| LifeSciBench | 59.9% | #3 of 5 | Lab model card | |
| MedChemBench (OpenAI) | 48.3% | #3 of 5 | Lab model card | |
| UK AISI unprompted sabotage evaluation | 0% | #3 of 5 | Lab model card | |
| MetagenomicsBench | 31.3% | #17 of 26 | Official leaderboard | |
| Prompt injection attacks in search and function calling | 91% | #2 of 3 | Lab model card | |
| First-Person Fairness evaluation | 0.98% | #7 of 10 | Lab model card | |
| Instruction hierarchy attacks | 99.9% | #5 of 7 | Lab model card | |
| DNA sequence design for transcription factor binding | 13.7% | #3 of 4 | Lab model card | |
| CoT-Control | 1.30% | #7 of 9 | Lab model card | |
| Prompt injection robustness | 96.2% | #7 of 7 | Lab model card | |
| UK AISI sabotage continuation evaluation | 1.00% | #4 of 4 | Lab model card | |
| FrontierCyber | 9.64% | #3 of 3 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 52.6% | #6 of 81 | Official leaderboard | |
| Conceptual Reasoning Index | 68.7 | #12 of 151 | Official leaderboard | |
| LMCA | 58.5% | #12 of 147 | Official leaderboard | |
| Political Science LLM Benchmark | 69.2% | #5 of 42 | Official leaderboard | |
| Vals Multimodal Index | 72.6% | #4 of 33 | Official leaderboard | |
| CompassRank | 80.8 | #2 of 14 | Official leaderboard | |
| CAIS Text Capabilities Index | 48.8 | #12 of 68 | Official leaderboard | |
| ACCoRD | 75.9 | #27 of 152 | Official leaderboard | |
| CAIS Vision Capabilities Index | 66.4 | #11 of 53 | Official leaderboard | |
| Artificial Analysis Intelligence Index v4.1 | 58.9 | #2 of 8 | Lab model card | |
| CAIS Risk Index | 44.5 | #23 of 67 | Official leaderboard | |
| Vals Science Index | 41.5% | #11 of 25 | Official leaderboard | |
| Xbench | 18.2 | #14 of 17 | Official leaderboard |
Its lab's numbers put it at 162.6, in line with its ECI
The Self-Reported Index fits one score to the 17 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 159.1 to 166.1. How we compute it.
- HealthBench Consensus 95.5
- GPQA Diamond 94.6
- BrowseComp 92.2
- Terminal-Bench 2.1 91.9
- OpenAI MRCR v2 91.5
- GraphWalks 90.7
- MMMU-Pro 84.6
- DeepSWE 72.7
- BenchCAD 70.6
- SWE-Bench Pro 64.6
- OSWorld 2.0 62.6
- HealthBench Professional 60.5
- Toolathlon 58
- HealthBench 57
- Agents' Last Exam 53.6
- HealthBench Hard 33.1
- AutomationBench 18.1
Its lab's numbers next to 13 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 |
|---|---|---|---|
| Agents' Last Exam | 53.6% Agents' Last Exam (Sol, xhigh, code mode) | 30.6% Leaderboard, xhigh, codex | +23 pts |
| AutomationBench | 18.1% AutomationBench (Sol, max) | 28.8% Leaderboard, max | −11 pts |
| OpenAI MRCR v2 | 91.5% OpenAI MRCR v2 8-needle 256K-512K 91.5% (512K-1M 73.8%) | 83.5% Leaderboard, 8-needle, 128K to 256K bin, max reasoning | +8.0 pts |
| Human Pathogen Capabilities Test | 68.4% SecureBio Human Pathogen Capabilities Test, strongest reported configuration (9.1.1.8) | 76.3% Leaderboard, max | −7.9 pts |
| World-Class Bio | 68.3% SecureBio World-Class Bio, strongest reported configuration (9.1.1.8) | 64.4% Leaderboard, max | +3.9 pts |
| GeneBench Pro | 28.7% GeneBench Pro (Sol, max) | 31.5% Leaderboard, pro | −2.8 pts |
| Molecular Biology Capabilities Test | 60% SecureBio Molecular Biology Capabilities Test, strongest reported configuration (9.1.1.8) | 61.1% Leaderboard, max | −1.1 pts |
| GPQA Diamond | 94.6% GPQA Diamond | 93.5% Epoch AI run, max | +1.1 pts |
| FrontierMath Tiers 1-3 v2 | 89% FrontierMath Tier 1-3 (v2) | 89.1% Leaderboard, max; task version 2.0.0, Epoch | Same |
| FrontierMath Tier 4 v2 | 83% FrontierMath Tier 4 (v2) | 82.9% Leaderboard, max; task version 2.0.0, Epoch | Same |
| DeepSWE | 72.7% DeepSWE v1.1 (Sol, max) | 72.7% Leaderboard, max, mini-swe-agent | Same |
| ARC-AGI-3 | 7.78% ARC-AGI-3 | 7.78% Leaderboard, max, standard | Same |
| Management Consulting Tasks (OpenAI) | 43.2% Management Consulting Tasks (Internal) | 43.2% Leaderboard | Same |
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.
Its system cards
Every result in the longer documents the lab published with it. 50 of them aren't in the launch post.
GPT-5.6 Preview System Card Jun 25, 2026
Health
Hallucination
Safety
- OpenAI Production Benchmarks and safe-completion evals
- Disallowed Content Evaluations with Representative Prompts
- Image input safety evaluations
- Destructive action avoidance 83
- User Confirmations During Computer Use
- Biology Model Refusal Evaluation
- OpenAI cyber safety evaluations (incl. Daybreak Blue)
Robustness
Bias
Alignment
Monitorability
Bio and chem risk
- Multimodal Troubleshooting Virology 55.5
- ProtocolQA Open-Ended 43.5
- Tacit Knowledge and Troubleshooting
- TroubleshootingBench 48
- AAV Capsid Packaging Prediction (OpenAI) 0.529
- Hard-negative protein binding prediction 7.6
- DNA sequence design for transcription factor binding 13.7
- Virology Capabilities Test (VCT) 53.5
- Molecular Biology Capabilities Test 60
- Human Pathogen Capabilities Test 68.4
- World-Class Bio 68.3
- ReproBAIT 85
Cyber risk
GPT-5.6 System Card Jul 9, 2026
Health
Hallucination
Safety
- OpenAI Production Benchmarks and safe-completion evals
- Disallowed Content Evaluations with Representative Prompts
- Image input safety evaluations
- Destructive action avoidance 83
- User Confirmations During Computer Use
- Biology Model Refusal Evaluation
- OpenAI cyber safety evaluations (incl. Daybreak Blue)
Robustness
Bias
Alignment
Monitorability
Bio and chem risk
- Multimodal Troubleshooting Virology 55.5
- ProtocolQA Open-Ended 43.5
- Tacit Knowledge and Troubleshooting
- TroubleshootingBench 48
- AAV Capsid Packaging Prediction (OpenAI) 0.529
- Hard-negative protein binding prediction 7.6
- DNA sequence design for transcription factor binding 13.7
- Virology Capabilities Test (VCT) 53.5
- Molecular Biology Capabilities Test 60
- Human Pathogen Capabilities Test 68.4
- World-Class Bio 68.3
- ReproBAIT 85
Cyber risk
GPT-5.6 – August Updates Aug 6, 2026
Health
Hallucination
Safety
Robustness
Bio and chem risk
Cyber risk
- Capture the Flag (Internal) 97.1
- CVE-Bench 94.1
- Cyber Range 83.3