FrontierMath: leaderboard
Epoch AI's expert-crafted math benchmark featuring original, unpublished problems that test genuine mathematical reasoning rather than memorization.
Metric: Accuracy (%, 285 private v2 problems). Source: epoch.ai. Source: epoch.ai. 117 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 93.68 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 93.68 | #7 (GPT-6.1 Sol) |
| 3 | Claude Opus 5.5 (Max) | 91.23 | #3 (Claude Opus 5.5) |
| 4 | Claude Fable 5.1 (Max) | 90.18 | #9 (Claude Fable 5.1) |
| 5 | GPT-6 Sol (Max) | 89.82 | #14 (GPT-6 Sol) |
| 6 | GPT-5.6 Sol (Max) | 89.12 | #19 (GPT-5.6 Sol) |
| 7 | Claude Sonnet 5.5 (Max) | 88.77 | #11 (Claude Sonnet 5.5) |
| 8 | GPT-5.5 Pro (xHigh) | 87.72 | #17 (GPT-5.5 Pro) |
| 9 | Claude Fable 5 (Max) | 87.02 | #16 (Claude Fable 5) |
| 10 | GPT-5.6 Terra (Max) | 85.96 | #41 (GPT-5.6 Terra) |
| 11 | Gemini 4 Argon (High) | 85.96 | #8 (Gemini 4 Argon) |
| 12 | Claude Opus 5 (Max) | 85.61 | #12 (Claude Opus 5) |
| 13 | GPT-5.5 (xHigh) | 85.26 | #31 (GPT-5.5) |
| 14 | GPT-5.4 Pro (xHigh) | 82.46 | #32 (GPT-5.4 Pro) |
| 15 | GPT-5.6 Luna (Max) | 82.11 | #68 (GPT-5.6 Luna) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=frontiermath · How It Works · Data refreshed daily, snapshot 2026-10-20.