Introducing GPT-6 Sol and Luna (OpenAI) 22 Sep 2026
GPT-6 Luna
GPT-6 Luna ranks first on 4 of the 163 benchmarks it has scores on.
OpenAI Released 22 Sep 2026
171
Benchmarks with a score
4
Ranked first, out of five or more models
7
In the top three
OpenAI Sep 2026
GPT-6 Luna
Closed weights
156.8
#28 of 271
Epoch Capabilities Index
- Terminal-Bench 4.0 16.4% #19 /23
- ZeroBench 20% #12 /66
- HealthBench Professional 60.8% #12 /25
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 | |
|---|---|---|---|---|
| FACTS Benchmark Suite | 62.3% | #9 of 40 | Official leaderboard | |
| LMArena (Elo) | 1,441 | #77 of 333 | Official leaderboard | |
| BullshitBench V2 | 59% | #32 of 122 | Official leaderboard | |
| FACTS Grounding v2 | 76.7% | #4 of 15 | Official leaderboard | |
| BullshitBench | 40% | #39 of 107 | Official leaderboard | |
| Global MMLU Lite | 92% | #18 of 43 | Official leaderboard | |
| FACTS Search V2 | 81.4% | #10 of 21 | Official leaderboard | |
| OpenAI user-flagged hallucination eval | 24.6% | #2 of 4 | Model card | |
| SimpleQA Verified | 41.4% | #46 of 84 | Epoch AI | |
| FACTS Parametric V2 | 57% | #11 of 20 | Official leaderboard | |
| Extended NYT Connections | 68.7% | #54 of 97 | Official leaderboard | |
| Factuality Heavy prompts | 5.60% | #2 of 3 | Model card | |
| High Stakes factuality prompts | 4.50% | #2 of 3 | Model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.9% | #22 of 202 | Epoch AI | |
| GPQA Diamond | 90.5% | #41 of 219 | Epoch AI | |
| CritPt | 19.4% | #32 of 160 | Epoch AI | |
| FrontierMath Tiers 1-3 v2 | 79% | #19 of 89 | Epoch AI | |
| ProofBench | 64% | #20 of 83 | Epoch AI | |
| DTBench | 83.5% | #49 of 178 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 56.1% | #20 of 71 | Epoch AI | |
| Verifying the Verifiers (Project APE CRED benchmark) | 87.5% | #9 of 23 | Official leaderboard | |
| ProofBench v1.1 | 64% | #20 of 51 | Official leaderboard | |
| ARC-AGI-2 | 59.3% | #38 of 95 | Official leaderboard | |
| ARC-AGI | 86.7% | #41 of 92 | Official leaderboard | |
| Prophet Arena | 85.1 | #10 of 18 | Official leaderboard | |
| BixBench3 | 36.9 | #14 of 20 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 37.2% | #20 of 27 | Official leaderboard | |
| MysteryMechanism | 19.4% | #25 of 29 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ALE-Bench | 1,577 | #22 of 121 | Epoch AI | |
| Vibe Code Bench | 81.7% | #25 of 115 | Official leaderboard | |
| WebDev Arena | 1,582 | #23 of 105 | Epoch AI | |
| Code Arena | 1,579 | #36 of 124 | Official leaderboard | |
| OpenDesign Arena | 83.1 | #6 of 19 | Official leaderboard | |
| Code Migration | 42.5% | #28 of 78 | Official leaderboard | |
| Brokk Power Ranking | 10.7 | #7 of 19 | Official leaderboard | |
| Image-to-WebDev Arena | 1,571 | #24 of 55 | Official leaderboard | |
| FrontierCode | 42.4% | #22 of 46 | Epoch AI | |
| IOI v2 (Vals) | 55.6% | #22 of 45 | Official leaderboard | |
| DeepSWE | 66.6% | #27 of 51 | Model card | |
| FrontierCode 2.0 | 21.9% | #16 of 26 | Official leaderboard | |
| Code Migration v1.1 | 46.7% | #18 of 29 | Official leaderboard | |
| WeirdML v3 | 7.87% | #16 of 25 | Official leaderboard | |
| Android Bench 2.0 | 10.7% | #9 of 11 | Official leaderboard | |
| MirrorCode | 26.7% | #23 of 28 | Epoch AI | |
| SWE-Marathon v1.1 | 28.1% | #15 of 17 | Official leaderboard | |
| FrontierCode 1.1 | 42.4% | #11 of 12 | Model card | |
| Next.js Agent Evals | 81% | #13 of 14 | Official leaderboard | |
| Internal Research Debugging Evaluation | 46.6% | #6 of 6 | Model card | |
| KernelGen 1P | 21.8% | #4 of 4 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LLM Chess | 1,219 | #14 of 124 | Official leaderboard | |
| LisanBench | 3863.7 | #30 of 133 | Official leaderboard | |
| LMArena Search Arena | 1,209 | #13 of 35 | Official leaderboard | |
| Agent Arena | 0.02 | #22 of 53 | Official leaderboard | |
| CivBench | 1,510 | #8 of 18 | Official leaderboard | |
| Chess Puzzles | 36% | #43 of 83 | Epoch AI | |
| ARC-AGI-3 | 0.59% | #15 of 24 | Official leaderboard | |
| OSWorld 2.0 | 52.7% | #13 of 19 | Model card | |
| Hermes Bench | 70.5% | #12 of 17 | Official leaderboard | |
| Game Arena Word Art | 254 | #25 of 35 | Official leaderboard | |
| Game Arena Word Association | 122 | #26 of 36 | Official leaderboard | |
| Game Arena Dark Hex | 137 | #28 of 38 | Official leaderboard | |
| Game Arena Checkers | 641 | #29 of 38 | Official leaderboard | |
| Game Arena Bargaining | 148 | #29 of 36 | Official leaderboard | |
| Game Arena Go | 476 | #29 of 36 | Official leaderboard | |
| Game Arena Reversi | 184 | #29 of 36 | Official leaderboard | |
| Terminal-Bench 4.0 | 16.4% | #19 of 23 | Official leaderboard | |
| Game Arena Dots and Boxes | 611 | #31 of 37 | Official leaderboard | |
| Game Arena Chess | 449 | #33 of 39 | Official leaderboard | |
| Kaggle Game Arena | 169 | #33 of 39 | Official leaderboard | |
| AutomationBench | 20.7% | #35 of 41 | Model card | |
| SpireClimb StS1 A20 Heart | 0% | #6 of 7 | Epoch AI | |
| Game Arena Clobber | 60 | #31 of 36 | Official leaderboard | |
| Game Arena Lines of Action | 319 | #31 of 36 | Official leaderboard | |
| Game Arena Coin Game | 174 | #32 of 35 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 22 | #33 of 36 | Official leaderboard | |
| Game Arena Four in a Row | 127 | #35 of 38 | Official leaderboard | |
| Mystery Game Puzzles | 7.00% | #75 of 80 | Epoch AI | |
| Game Arena Nine Men's Morris | 0 | #36 of 36 | Official leaderboard | |
| StarSkirmish Bench | 13.1% | #9 of 9 | Official leaderboard | |
| SpireClimb | 0% | #7 of 7 | Epoch AI | |
| SpireClimb StS2 A10 | 0% | #7 of 7 | Epoch AI |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PolicyBench | 91.5% | #7 of 50 | Official leaderboard | |
| Agents' Last Exam | 50.9% | #4 of 25 | Model card | |
| Excel Modeling Benchmark | 68.5% | #16 of 76 | Official leaderboard | |
| HealthBench Consensus | 95.9% | #5 of 17 | Model card | |
| MedCode | 44.7% | #39 of 96 | Official leaderboard | |
| MedScribe | 83.7% | #41 of 98 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 70% | #16 of 37 | Official leaderboard | |
| HealthBench Professional | 60.8% | #12 of 25 | Model card | |
| Harvey LAB (Legal Agent Benchmark) | 2.92% | #44 of 80 | Official leaderboard | |
| Vals Finance Agent v2 | 49.9% | #45 of 80 | Official leaderboard | |
| TaxCalcBench v2 | 36% | #15 of 26 | Official leaderboard | |
| Legal Research Bench | 30.3% | #52 of 88 | Official leaderboard | |
| PRBench | 42.9% | #27 of 44 | Official leaderboard | |
| Tax Agent Bench | 19.9% | #52 of 79 | Official leaderboard | |
| CAD Arena | 0.30 | #8 of 12 | Official leaderboard | |
| APEX-Agents | 44.3% | #40 of 56 | Epoch AI | |
| HealthBench Hard | 31.4% | #15 of 21 | Model card | |
| Public Benefits Bench | 57.6% | #38 of 50 | Official leaderboard | |
| Vals Finance Index | 53% | #27 of 35 | Official leaderboard | |
| Vals Index v2.2 | 52.3% | #25 of 32 | Official leaderboard | |
| Clipping Bench | $374 | #15 of 19 | Official leaderboard | |
| Vals Finance Agent v2.1 | 56.8% | #28 of 35 | Official leaderboard | |
| RCM-Bench-1 | 85.8% | #16 of 20 | Official leaderboard | |
| Immigration Bench | 62.5% | #20 of 24 | Official leaderboard | |
| EEBench | 31.1% | #22 of 26 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 4.17% | #28 of 33 | Official leaderboard | |
| Vals Legal Index | 62.2% | #28 of 33 | Official leaderboard | |
| HealthBench | 54.5% | #26 of 30 | Model card | |
| Thrive Holdings helpdesk ticket resolution benchmark | 34.6% | #13 of 15 | Official leaderboard | |
| Short Form Bench | $2,107 | #8 of 9 | Official leaderboard | |
| Holdings-IndividualTaxBench | 67.6% | #12 of 13 | Official leaderboard | |
| Legora BAR | 88.0 | #17 of 17 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ZeroBench | 20% | #12 of 66 | Official leaderboard | |
| VoxelBench | 1,809 | #21 of 56 | Official leaderboard | |
| Vision Arena | 1,240 | #56 of 136 | Official leaderboard | |
| Furniture Assembly Benchmark | 44.2% | #15 of 35 | Epoch AI | |
| GDP.pdf | 23% | #19 of 39 | Epoch AI | |
| ParseBench | 64.3% | #21 of 42 | Official leaderboard | |
| Blueprint-Bench 2 | 31.2% | #18 of 35 | Official leaderboard | |
| Code2Video Benchmark | 1,469 | #14 of 19 | Official leaderboard | |
| Humanity's Sixth Sense | 21% | #29 of 29 | Official leaderboard | |
| 1H-VideoQA | 57.4% | #20 of 20 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Tacit Knowledge and Troubleshooting | 94% | #1 of 29 | Model card | |
| FORTRESS | 10.8% | #5 of 66 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 76.3% | #5 of 43 | Official leaderboard | |
| Phage-plasmid Co-evolution | 10.7 | #1 of 5 | Model card | |
| Respecting Warnings | 11.8% | #1 of 5 | Model card | |
| Unintended Engagement with External Agent Messages | 0% | #1 of 5 | Model card | |
| AAV Capsid Packaging Prediction (OpenAI) | 0.53 | #2 of 6 | Model card | |
| TroubleshootingBench | 41.5% | #7 of 19 | Model card | |
| Declining to Exploit a Honeypot During Difficult ExploitGym Problems | 0% | #2 of 5 | Model card | |
| BioSecBench-Refusal | 37.5% | #10 of 22 | Official leaderboard | |
| DeepsecBench | 21.0 | #13 of 28 | Official leaderboard | |
| Coding Deception | 2.81% | #3 of 6 | Model card | |
| ProtocolQA Open-Ended | 34.9% | #14 of 25 | Model card | |
| OncologyBench | 16% | #17 of 30 | Official leaderboard | |
| Broken Search Tool | 7.83% | #4 of 7 | Model card | |
| Instruction hierarchy attacks | 100% | #4 of 7 | Model card | |
| Prompt injection robustness | 98.6% | #4 of 7 | Model card | |
| SRE-Bench | 2.67% | #21 of 33 | Official leaderboard | |
| TxBench-Oligo-Discovery | 33.6% | #18 of 28 | Official leaderboard | |
| Blocker Deception | 11.1% | #2 of 3 | Model card | |
| TxBench-Antibody-Discovery | 34% | #19 of 28 | Official leaderboard | |
| MetagenomicsBench | 31% | #18 of 26 | Official leaderboard | |
| BioSecBench-Surveillance | 41.5% | #19 of 27 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 44.3% | #20 of 28 | Official leaderboard | |
| scBench-Long | 11.9% | #20 of 27 | Official leaderboard | |
| MentalHealthBench | 51.7% | #3 of 4 | Model card | |
| Sandbox Bench | 4.50% | #3 of 4 | Model card | |
| Multimodal Troubleshooting Virology | 49.1% | #20 of 26 | Model card | |
| SpatialBench-Long | 23.2% | #21 of 27 | Official leaderboard | |
| ExploitBench | 46.2% | #15 of 19 | Model card | |
| Respecting Auto-Review | 0% | #4 of 5 | Model card | |
| DistressBench | 66.7% | #34 of 42 | Official leaderboard | |
| EpiBench | 19.2% | #22 of 27 | Official leaderboard | |
| VariantBench | 27.7% | #23 of 27 | Official leaderboard | |
| SEC-bench Pro | 51.8% | #9 of 10 | Model card | |
| BioSecBench-Function | 27.3% | #25 of 27 | Official leaderboard | |
| ExploitGym | 11.6% | #13 of 14 | Model card | |
| Coronavirus-ACE2 Cell-Entry Screen | 0.37 | #5 of 5 | Model card | |
| SHP2 Protein Function Prediction | 0.21 | #5 of 5 | Model card | |
| ExploitBench - Internal Port | 0% | #4 of 4 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 72.1 | #34 of 150 | Official leaderboard | |
| LMCA | 44.5% | #40 of 139 | Epoch AI | |
| Conceptual Reasoning Index | 57.8 | #44 of 149 | Official leaderboard | |
| SAGE (Vals) | 48.1% | #28 of 81 | Official leaderboard | |
| LiveBench | 64.6% | #17 of 28 | Official leaderboard | |
| SAGE v1.1 (Vals) | 83.4% | #20 of 28 | Official leaderboard | |
| Hermes Index | 33.9 | #14 of 17 | Official leaderboard | |
| Vals Science Index | 28.4% | #20 of 23 | Official leaderboard |
OpenAI's models over time
OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.
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
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