FrontierMath - Tier 4 (v2): leaderboard
One item is a research-level mathematics problem from Tier 4, the 43-problem expansion set holding the most difficult FrontierMath problems. The model reasons, may run Python in a stateless sandbox with a fixed library list, and submits an answer() function whose returned object is checked automatically, one point per correct problem. This board covers the 41 private v2 problems, higher is better.
Metric: Accuracy (%, 41 private v2 problems). Source: epoch.ai. Saturation forecast: Around January 2027. 76 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (Max) | 100 | #7 (GPT-6.1 Sol) |
| 2 | GPT-6 Astra (Max) | 97.6 | #6 (GPT-6 Astra) |
| 3 | GPT-6 Astra (xHigh) | 97.6 | #6 (GPT-6 Astra) |
| 4 | GPT-6 Astra (High) | 97.6 | #6 (GPT-6 Astra) |
| 5 | GPT-6 Astra (Medium) | 97.56 | #6 (GPT-6 Astra) |
| 6 | Claude Opus 5.5 (Max) | 95 | #3 (Claude Opus 5.5) |
| 7 | Claude Fable 5 (Max) | 90.2 | #16 (Claude Fable 5) |
| 8 | GPT-6 Sol (Max) | 90 | #14 (GPT-6 Sol) |
| 9 | Claude Fable 5.1 (Max) | 87.8 | #9 (Claude Fable 5.1) |
| 10 | GPT-6 Astra (Low) | 87.8 | #6 (GPT-6 Astra) |
| 11 | GPT-5.6 Sol (Max) | 82.93 | #19 (GPT-5.6 Sol) |
| 12 | Claude Sonnet 5.5 (Max) | 80.49 | #11 (Claude Sonnet 5.5) |
| 13 | Gemini 4 Argon (High) | 78.05 | #8 (Gemini 4 Argon) |
| 14 | GPT-5.5 Pro (xHigh) | 78.05 | #17 (GPT-5.5 Pro) |
| 15 | Claude Opus 5 (Max) | 73.17 | #12 (Claude Opus 5) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=frontiermath-tier-4-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.