Introducing GPT-6 Sol and Luna (OpenAI) Sep 22, 2026
GPT-6 Luna
GPT-6 Luna ranks first on 4 of the 252 benchmarks with five or more models scored.
260
Benchmarks with a score
4
Ranked first, out of five or more models
8
In the top three
OpenAI Sep 2026
GPT-6 Luna
Closed weights
156.4
#36 of 256
Epoch Capabilities Index
Self-Reported Index 157.3 #48 of 404
- AutomationBench 20.7% #17 /42
- Agents' Last Exam 25% #16 /31
- Terminal-Bench 4.0 16.4% #28 /40
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 | |
|---|---|---|---|---|
| HLE Extended | 53.3% | #7 of 31 | Official leaderboard | |
| LMArena (Elo) | 1,441 | #78 of 335 | Official leaderboard | |
| FACTS Grounding v2 | 76.7% | #7 of 25 | Official leaderboard | |
| BullshitBench V2 | 59% | #37 of 129 | Official leaderboard | |
| StudyArena Text | 1,243 | #30 of 85 | Official leaderboard | |
| BullshitBench | 40% | #39 of 107 | Official leaderboard | |
| AA-LCR Extended | 68.3% | #18 of 49 | Official leaderboard | |
| Global MMLU Lite | 92% | #18 of 43 | Official leaderboard | |
| FACTS Search V2 | 81.4% | #10 of 21 | Official leaderboard | |
| Internal adversarially-designed factuality evaluations | 7.60% | #1 of 2 | Lab model card | |
| DesignArena ASCII Art | 1,161 | #39 of 77 | Official leaderboard | |
| FACTS Parametric V2 | 57% | #14 of 27 | Official leaderboard | |
| FACTS Benchmark Suite | 64.8% | #12 of 23 | Official leaderboard | |
| Opening Lines Arena | 1,515 | #8 of 15 | Official leaderboard | |
| SimpleQA Verified | 41.4% | #51 of 94 | Epoch AI run | |
| Extended NYT Connections | 68.7% | #60 of 103 | Official leaderboard | |
| Factuality Heavy prompts | 5.60% | #2 of 3 | Lab model card | |
| High Stakes factuality prompts | 4.50% | #2 of 3 | Lab model card | |
| Joke Arena | 1,486 | #18 of 25 | Official leaderboard | |
| OpenAI user-flagged hallucination eval | 18.7% | #3 of 4 | Lab model card | |
| ELI5 Arena | 1,422 | #14 of 18 | Official leaderboard | |
| FACTS Multimodal | 38.7% | #21 of 23 | Official leaderboard | |
| Corp Speak Arena | 1,218 | #22 of 22 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.9% | #27 of 212 | Epoch AI run | |
| GPQA Diamond | 90.5% | #45 of 229 | Epoch AI run | |
| FrontierMath Tiers 1-3 v2 | 79% | #20 of 86 | Official leaderboard | |
| DTBench | 90.1% | #51 of 180 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 56.1% | #20 of 68 | Official leaderboard | |
| PHYSICS-Verified | 82.6% | #7 of 19 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 87.5% | #9 of 23 | Official leaderboard | |
| ProofBench v1.1 | 64% | #21 of 52 | Official leaderboard | |
| ARC-AGI-2 | 59.3% | #40 of 98 | Official leaderboard | |
| ARC-AGI | 86.7% | #43 of 95 | Official leaderboard | |
| Prophet Arena | 85.6 | #13 of 24 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 53.9% | #17 of 26 | Official leaderboard | |
| BixBench3 | 36.9 | #14 of 20 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 37.2% | #20 of 27 | Official leaderboard | |
| MysteryMechanism | 19.4% | #27 of 31 | Official leaderboard | |
| BioMysteryBench | 61.5% | #27 of 30 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DesignArena UI Component | 1,292 | #26 of 139 | Official leaderboard | |
| WebDev Arena | 1,582 | #23 of 105 | arena.ai, via Epoch AI | |
| Vibe Code Bench | 81.7% | #28 of 120 | Official leaderboard | |
| DesignArena SVG | 1,238 | #25 of 103 | Official leaderboard | |
| DesignArena Website | 1,269 | #37 of 151 | Official leaderboard | |
| ALE-Bench | 1,791 | #36 of 126 | Official leaderboard | |
| DesignArena Code | 1,262 | #44 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,254 | #44 of 143 | Official leaderboard | |
| Code Arena | 1,577 | #40 of 128 | Official leaderboard | |
| OpenDesign Arena | 83.1 | #7 of 21 | Official leaderboard | |
| Code Migration | 42.5% | #28 of 78 | Official leaderboard | |
| DAXBench v2 | 83% | #17 of 47 | Official leaderboard | |
| Brokk Power Ranking | 10.7 | #7 of 19 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,181 | #20 of 46 | Official leaderboard | |
| Image-to-WebDev Arena | 1,566 | #24 of 55 | Official leaderboard | |
| DesignArena 3D Design | 1,198 | #60 of 136 | Official leaderboard | |
| IOI v2 (Vals) | 55.6% | #22 of 45 | Official leaderboard | |
| DesignArena Data Visualization | 1,209 | #61 of 118 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 15.3% | #20 of 38 | Official leaderboard | |
| DeepSWE | 66.6% | #30 of 56 | Lab model card | |
| DesignArena Full Stack | 1,140 | #29 of 49 | Official leaderboard | |
| FrontierCode 2.0 | 22.7% | #19 of 32 | Official leaderboard | |
| APEX-SWE | 38.8% | #39 of 65 | Official leaderboard | |
| DesignArena Agentic Overall | 1,146 | #46 of 74 | Official leaderboard | |
| Code Migration v1.1 | 46.7% | #19 of 30 | Official leaderboard | |
| DesignArena Frontend Web App | 1,162 | #34 of 53 | Official leaderboard | |
| DesignArena Image to HTML | 1,185 | #33 of 51 | Official leaderboard | |
| WeirdML v3 | 7.87% | #18 of 27 | Official leaderboard | |
| SWE-bench Verified Extended | 39.3% | #31 of 44 | Official leaderboard | |
| Next.js Agent Evals | 86% | #10 of 14 | Official leaderboard | |
| SWE-Together | 59.6% | #16 of 22 | Official leaderboard | |
| MacroscopeBench | 72.5% | #12 of 16 | Official leaderboard | |
| DesignArena React Native | 1,119 | #43 of 56 | Official leaderboard | |
| SWE Atlas v1.1 | 50.8% | #22 of 28 | Official leaderboard | |
| Real-SWE 1.1 | 10.4% | #12 of 15 | Official leaderboard | |
| Floorplan Arena | 1,380 | #9 of 11 | Official leaderboard | |
| MirrorCode | 26.7% | #32 of 39 | Epoch AI run | |
| FDE-Gym | -0.12 | #10 of 12 | Official leaderboard | |
| Chart Arena | 1,415 | #13 of 15 | Official leaderboard | |
| Android Bench 2.0 | 10.7% | #15 of 17 | Official leaderboard | |
| SWE-Marathon v1.1 | 28.1% | #18 of 20 | Official leaderboard | |
| FrontierCode 1.1 | 42.4% | #12 of 13 | Lab model card | |
| Internal Research Debugging Evaluation | 46.6% | #6 of 6 | Lab model card | |
| KernelGen 1P | 21.8% | #4 of 4 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LLM Chess | 1,219 | #15 of 123 | Official leaderboard | |
| BrowseComp Extended | 47.2% | #9 of 42 | Official leaderboard | |
| LisanBench | 3863.7 | #30 of 133 | Official leaderboard | |
| LMArena Search Arena | 1,211 | #12 of 37 | Official leaderboard | |
| o11y-bench | 84.4% | #8 of 22 | Official leaderboard | |
| AutomationBench | 20.7% | #17 of 42 | Official leaderboard | |
| RuneBench | 5.13 | #31 of 76 | Official leaderboard | |
| CivBench | 1,510 | #8 of 19 | Official leaderboard | |
| Agent Arena | 1.56% | #26 of 56 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 33.3% | #22 of 44 | Official leaderboard | |
| AdaExploreBench | 51.5% | #4 of 8 | Official leaderboard | |
| Chess Puzzles | 36% | #47 of 93 | Epoch AI run | |
| ARC-AGI-3 | 0.59% | #18 of 27 | Official leaderboard | |
| OSWorld 2.0 | 52.7% | #13 of 19 | Lab model card | |
| Terminal-Bench 4.0 | 16.4% | #28 of 40 | Official leaderboard | |
| Hermes Bench | 70.5% | #12 of 17 | Official leaderboard | |
| Terminal-Bench 4.1 | 15.2% | #31 of 43 | Official leaderboard | |
| Game Arena Word Art | 254 | #29 of 39 | Official leaderboard | |
| Game Arena Word Association | 117 | #30 of 40 | Official leaderboard | |
| DuelLab GameBench 2 | 55.9 | #43 of 57 | Official leaderboard | |
| Game Arena Dark Hex | 137 | #31 of 41 | Official leaderboard | |
| Game Arena Checkers | 639 | #32 of 41 | Official leaderboard | |
| Gym-Anything-RT | 1.00% | #4 of 5 | Official leaderboard | |
| Web Context (Nimble) | 67.9% | #9 of 11 | Official leaderboard | |
| Game Arena Bargaining | 142 | #32 of 39 | Official leaderboard | |
| Game Arena Go | 474 | #32 of 39 | Official leaderboard | |
| Game Arena Reversi | 181 | #32 of 39 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 43.4% | #15 of 18 | Official leaderboard | |
| Game Arena Dots and Boxes | 583 | #34 of 40 | Official leaderboard | |
| Game Arena Chess | 450 | #36 of 42 | Official leaderboard | |
| Kaggle Game Arena | 171 | #36 of 42 | Official leaderboard | |
| Cresta-Bench | 32.7% | #12 of 14 | Official leaderboard | |
| Game Arena Clobber | 66 | #34 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 196 | #34 of 39 | Official leaderboard | |
| BU Bench V2.1 | 69.2 | #14 of 16 | Official leaderboard | |
| SpireClimb StS1 A20 Heart | 0% | #8 of 9 | Official leaderboard | |
| Game Arena Coin Game | 177 | #36 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 25 | #36 of 39 | Official leaderboard | |
| Game Arena Four in a Row | 128 | #40 of 43 | Official leaderboard | |
| Mystery Game Puzzles | 7.00% | #79 of 84 | Epoch AI run | |
| Game Arena Nine Men's Morris | 0 | #39 of 39 | Official leaderboard | |
| SpireClimb | 0% | #9 of 9 | Epoch AI run | |
| SpireClimb StS2 A10 | 0% | #9 of 9 | Official leaderboard | |
| StarSkirmish Bench | 13.1% | #9 of 9 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PolicyBench | 91.5% | #7 of 50 | Official leaderboard | |
| Excel Modeling Benchmark | 68.5% | #16 of 76 | Official leaderboard | |
| HealthBench Consensus | 95.9% | #5 of 17 | Lab model card | |
| Vals Index v2 | 58.5% | #20 of 65 | Official leaderboard | |
| MedCode | 44.7% | #39 of 96 | Official leaderboard | |
| MedScribe | 83.7% | #41 of 97 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 70% | #19 of 42 | Official leaderboard | |
| HW-Bench | 30% | #7 of 15 | Official leaderboard | |
| HealthBench Professional | 60.8% | #13 of 26 | Lab model card | |
| Agents' Last Exam | 25% | #16 of 31 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 2.92% | #44 of 80 | Official leaderboard | |
| Vals Finance Agent v2 | 49.9% | #45 of 80 | Official leaderboard | |
| DAYJOB: Finance | 6.80% | #22 of 39 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 43.1% | #29 of 51 | Official leaderboard | |
| Vals Index v2.1 | 51.2% | #28 of 49 | Official leaderboard | |
| FoodTruck Bench | $17,211 | #34 of 59 | Official leaderboard | |
| Tuesday Work Index | 45.3 | #26 of 44 | Official leaderboard | |
| GDPx | 5.11% | #12 of 20 | Official leaderboard | |
| Legal Research Bench | 30.3% | #57 of 93 | Official leaderboard | |
| APEX-1 | 62% | #44 of 71 | Official leaderboard | |
| TaxCalcBench v2 | 36% | #20 of 32 | Official leaderboard | |
| PRBench | 42.9% | #31 of 48 | Official leaderboard | |
| Bedside Bench | 87.2% | #11 of 17 | Official leaderboard | |
| CAD Arena | 0.30 | #8 of 12 | Official leaderboard | |
| Tax Agent Bench | 19.9% | #57 of 84 | Official leaderboard | |
| HealthBench Hard | 31.4% | #15 of 21 | Lab model card | |
| APEX-Agents | 44.3% | #45 of 61 | Collected by Epoch AI, source not named | |
| GDP.xlsx | 14.9% | #15 of 20 | Official leaderboard | |
| APEX-Agents 1.1 | 59% | #49 of 65 | Official leaderboard | |
| APEX-Accounting | 53.9% | #37 of 49 | Official leaderboard | |
| Public Benefits Bench | 57.6% | #38 of 50 | Official leaderboard | |
| DeepFinance | 26.4% | #18 of 23 | Official leaderboard | |
| DAYJOB: Healthcare | 1.20% | #29 of 37 | Official leaderboard | |
| Clipping Bench | $374 | #15 of 19 | Official leaderboard | |
| Vals Finance Index | 53% | #32 of 40 | Official leaderboard | |
| RCM-Bench-1 | 85.8% | #16 of 20 | Official leaderboard | |
| Vals Index v2.2 | 52.3% | #30 of 37 | Official leaderboard | |
| YC-Bench | $331,093 | #48 of 59 | Official leaderboard | |
| Vals Finance Agent v2.1 | 56.8% | #33 of 40 | Official leaderboard | |
| DesignArena PDF | 1,120 | #31 of 37 | Official leaderboard | |
| Immigration Bench | 62.5% | #21 of 25 | Official leaderboard | |
| EEBench | 31.1% | #26 of 30 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 34.6% | #13 of 15 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 4.17% | #33 of 38 | Official leaderboard | |
| Vals Legal Index | 62.2% | #33 of 38 | Official leaderboard | |
| HealthBench | 54.5% | #27 of 31 | Lab model card | |
| Short Form Bench | $2,107 | #8 of 9 | Official leaderboard | |
| MedXpertQA MM Extended | 46.4% | #37 of 41 | Official leaderboard | |
| Holdings-IndividualTaxBench | 67.6% | #12 of 13 | Official leaderboard | |
| GDPval Extended | 0% | #43 of 43 | Official leaderboard | |
| DesignArena Spreadsheet | 947 | #35 of 35 | Official leaderboard | |
| Legora BAR | 88.0 | #17 of 17 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BVB | 70.3 | #11 of 45 | Official leaderboard | |
| ZeroBench | 20% | #22 of 76 | Official leaderboard | |
| ExtractBench | 91% | #9 of 27 | Official leaderboard | |
| VoxelBench | 1,818 | #20 of 56 | Official leaderboard | |
| Vision Arena | 1,235 | #56 of 136 | Official leaderboard | |
| Chartography | 29.1% | #18 of 43 | Official leaderboard | |
| Furniture Assembly Benchmark | 44.2% | #16 of 35 | Official leaderboard | |
| ParseBench | 64.1% | #21 of 44 | Official leaderboard | |
| MMMU-Pro Extended | 34.8% | #19 of 38 | Official leaderboard | |
| SplatBench | 8.60 | #4 of 8 | Official leaderboard | |
| Code2Video-Compass | 7.40 | #8 of 15 | Official leaderboard | |
| Blueprint-Bench 2 | 31.2% | #21 of 38 | Official leaderboard | |
| GlobeBench | 17.8% | #12 of 19 | Official leaderboard | |
| Frontier3DBench | 48.6 | #8 of 11 | Official leaderboard | |
| Code2Video Benchmark | 1,469 | #14 of 19 | Official leaderboard | |
| CharXiv Extended | 76.4% | #30 of 38 | Official leaderboard | |
| vCX Benchmark | 80.7% | #36 of 42 | Official leaderboard | |
| DesignArena Code to Video | 1,028 | #27 of 29 | Official leaderboard | |
| Humanity's Sixth Sense | 21% | #34 of 34 | Official leaderboard | |
| 1H-VideoQA | 57.4% | #20 of 20 | Official leaderboard | |
| ExtractBench Pro | 0% | #7 of 7 | Official leaderboard | |
| ShadowBench | 35.1 | #6 of 6 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Tacit Knowledge and Troubleshooting | 94% | #1 of 30 | Lab model card | |
| FORTRESS | 10.8% | #6 of 73 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 76.3% | #5 of 44 | Official leaderboard | |
| Respecting Warnings | 11.8% | #1 of 7 | Lab model card | |
| Phage-plasmid Co-evolution | 10.7 | #1 of 6 | Lab model card | |
| Unintended Engagement with External Agent Messages | 0% | #1 of 6 | Lab model card | |
| BioTIER Refuse | 79.5% | #18 of 69 | Official leaderboard | |
| AAV Capsid Packaging Prediction (OpenAI) | 0.53 | #2 of 7 | Lab model card | |
| Declining to Exploit a Honeypot During Difficult ExploitGym Problems | 0% | #2 of 7 | Lab model card | |
| nocot-bench (NCRI) | 77.3 | #66 of 185 | Official leaderboard | |
| TroubleshootingBench | 41.5% | #8 of 20 | Lab model card | |
| BioSecBench-Refusal | 37.5% | #10 of 22 | Official leaderboard | |
| DeepsecBench | 21.0 | #14 of 29 | Official leaderboard | |
| Coding Deception | 2.81% | #3 of 6 | Lab model card | |
| OncologyBench | 16% | #17 of 30 | Official leaderboard | |
| Instruction hierarchy attacks | 99.97% | #4 of 7 | Lab model card | |
| Prompt injection robustness | 98.6% | #4 of 7 | Lab model card | |
| Respecting Auto-Review | 0% | #4 of 7 | Lab model card | |
| Suicide Memory Bench | 0.66 | #7 of 12 | Official leaderboard | |
| ProtocolQA Open-Ended | 34.9% | #16 of 27 | Lab model card | |
| Blocker Deception | 11.1% | #3 of 5 | Lab model card | |
| SRE-Bench | 2.67% | #22 of 35 | Official leaderboard | |
| TxBench-Oligo-Discovery | 33.6% | #20 of 30 | Official leaderboard | |
| Broken Search Tool | 7.83% | #6 of 9 | Lab model card | |
| MetagenomicsBench | 31% | #18 of 26 | Official leaderboard | |
| BioTIER Permit | 98.9% | #48 of 69 | Official leaderboard | |
| TxBench-Antibody-Discovery | 34% | #21 of 30 | Official leaderboard | |
| BioSecBench-Surveillance | 41.5% | #19 of 27 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 44.3% | #22 of 30 | Official leaderboard | |
| scBench-Long | 11.9% | #20 of 27 | Official leaderboard | |
| MentalHealthBench | 51.7% | #3 of 4 | Lab model card | |
| Sandbox Bench | 4.50% | #3 of 4 | Lab model card | |
| ExploitBench | 46.2% | #16 of 21 | Lab model card | |
| Multimodal Troubleshooting Virology | 49.1% | #21 of 27 | Lab model card | |
| SpatialBench-Long | 23.2% | #21 of 27 | Official leaderboard | |
| dfbench | 45.7% | #14 of 18 | Official leaderboard | |
| DistressBench | 66.7% | #34 of 42 | Official leaderboard | |
| EpiBench | 19.2% | #22 of 27 | Official leaderboard | |
| BioSimBench | 0.12 | #10 of 12 | Official leaderboard | |
| IPI Bench | 31.8% | #5 of 6 | Official leaderboard | |
| VariantBench | 27.7% | #23 of 27 | Official leaderboard | |
| SEC-bench Pro | 51.8% | #10 of 11 | Lab model card | |
| BioSecBench-Function | 27.3% | #25 of 27 | Official leaderboard | |
| ExploitGym | 11.6% | #14 of 15 | Lab model card | |
| SecIT Bench | 58% | #26 of 26 | Official leaderboard | |
| Coronavirus-ACE2 Cell-Entry Screen | 0.37 | #6 of 6 | Lab model card | |
| SHP2 Protein Function Prediction | 0.21 | #6 of 6 | Lab model card | |
| ExploitBench - Internal Port | 0% | #5 of 5 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 72.1 | #37 of 152 | Official leaderboard | |
| JevBench | 98.4% | #1 of 4 | Official leaderboard | |
| LMCA | 44.5% | #44 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 57.8 | #46 of 151 | Official leaderboard | |
| SAGE (Vals) | 48.1% | #28 of 81 | Official leaderboard | |
| Political Science LLM Benchmark | 66.6% | #21 of 42 | Official leaderboard | |
| Xbench | 28.9 | #10 of 17 | Official leaderboard | |
| SAGE v1.1 (Vals) | 83.4% | #21 of 29 | Official leaderboard | |
| LiveBench | 60.4% | #23 of 31 | Official leaderboard | |
| Hermes Index | 33.9 | #14 of 17 | Official leaderboard | |
| Vals Science Index | 28.4% | #22 of 25 | Official leaderboard |
Its lab's numbers put it at 157.3, in line with its ECI
The Self-Reported Index fits one score to the 9 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 152.6 to 162.1. How we compute it.
- HealthBench Consensus 95.9
- DeepSWE 66.6
- HealthBench Professional 60.8
- HealthBench 54.5
- OSWorld 2.0 52.7
- Agents' Last Exam 50.9
- FrontierCode 1.1 42.4
- HealthBench Hard 31.4
- AutomationBench 20.7
Its lab's numbers next to 2 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 | 50.9% Agents' Last Exam (GPT-6 Luna max) | 25% Leaderboard, max, codex | +26 pts |
| AutomationBench | 20.7% AutomationBench 1.0.6 (GPT-6 Luna max) | 20.7% Leaderboard, max | 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. 33 of them aren't in the launch post.
Health
Hallucination
Safety
Robustness
Alignment
Bio and chem risk
Cyber risk
- ExploitBench 43.4
- ExploitBench - Internal Port 0
- SEC-bench Pro 34.2
- ExploitGym 11.6
- Sandbox Bench 4.5
AI R&D risk
GPT-6 Sol and GPT-6 Luna: October 2026 update Oct 7, 2026
Health
Hallucination
Safety
Robustness
Alignment
- Respecting Auto-Review 0
- Respecting Warnings 15.9
- Blocker Deception 13.3
- Broken Search Tool 9.7
Bio and chem risk
Cyber risk
- ExploitBench 44.7
- SEC-bench Pro 48.5
Health
Hallucination
Safety
Robustness
Alignment
- Respecting Auto-Review 0
- Respecting Warnings 11.8
- Blocker Deception 11.1
- Broken Search Tool 7.83
Bio and chem risk
Cyber risk
- ExploitBench 46.2
- SEC-bench Pro 51.8