FrontierMath - Tier 4 (v2) — leaderboard
Metric: Accuracy (%, 41 private v2 problems). Source: epoch.ai. 40 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 (Max) | 87.8 |
| 2 | GPT-5.6 Sol (Max) | 82.93 |
| 3 | GPT-5.5 Pro (xHigh) | 78.05 |
| 4 | GPT-5.5 (xHigh) | 72.5 |
| 5 | GPT-5.6 Terra (Max) | 70.73 |
| 6 | GPT-5.6 Luna (Max) | 60.98 |
| 7 | GPT-5.4 Pro (xHigh) | 58.54 |
| 8 | Claude Opus 4.8 (Max) | 56.1 |
| 9 | GPT-5.4 (xHigh) | 49 |
| 10 | Kimi K3 (Max) | 39.02 |
| 11 | Qwen 3.7 Max | 34.15 |
| 12 | Claude Opus 4.7 (Max) | 31.71 |
| 13 | GPT-5.2 (xHigh) | 31.7 |
| 14 | GLM-5.2 (Max) | 29.27 |
| 15 | Claude Sonnet 5 (Max) | 29.27 |
Interactive version: theaggregate.ai/benchmark?slug=frontiermath-tier-4-v2 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.