GPT-6 Luna (Max): benchmark results
Provider: OpenAI. Released 2026-09-22. Access: API.
Aggregate Rating 1805 ± 8, rank #130 of 2178 rated models, from 316 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenAI GPT-6.1 Terra System Card - External Agent Messages (Corrected Paths October 2026) - Engagement | 0 | Engagement given discovery (%; 80 samples; includes refusals | 100 |
| OpenAI GPT-6.1 Terra System Card - Successful Warning Circumvention (October 2026) | 17.4 | Successful circumvention rate (%; without safeguards) | 100 |
| AA-AnalystAgent - Pass@K | 85 | Pass@K (%) | 98 |
| AA Long Context Reasoning | 83.33 | Accuracy (%) | 96.2 |
| OTIS Mock AIME 2024-25 | 98.89 | Accuracy (%) | 92.2 |
| OpenAI GPT-6.1 Terra System Card - External Agent Messages (Corrected Paths October 2026) - Unauthorized Action | 0 | Specified unauthorized action given discovery (%; 80 samples | 91.7 |
| Vals AI CyberBench | 76.25 | Accuracy (%) | 91.1 |
| MLCR-AA - Accuracy | 88.89 | Accuracy (%) | 90.3 |
| ObviousBench | 99.31 | Answer pass³ (%) | 89.9 |
| AA CritPt | 19.43 | Accuracy (%) | 88.1 |
| FrontierMath - Tiers 1-3 (v2) | 78.95 | Accuracy (%, 285 private v2 problems) | 87.8 |
| AA-AnalystAgent - Pass@1 | 65.75 | Pass@1 (%) | 87.3 |
Interactive version: theaggregate.ai/model?slug=gpt-6-luna-max · How It Works · Data refreshed daily, snapshot 2026-11-07.