FrontierMath - Tier 4 (v2): leaderboard
Metric: Accuracy (%, 41 private v2 problems). Source: epoch.ai. 60 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 (Medium) | 97.56 |
| 2 | GPT-6 (Max) | 95.12 |
| 3 | GPT-6 (xHigh) | 95.12 |
| 4 | GPT-6 (High) | 95.12 |
| 5 | Claude Fable 5 (Max) | 87.8 |
| 6 | Claude Fable 5.1 (Max) | 87.8 |
| 7 | GPT-6 (Low) | 87.8 |
| 8 | GPT-5.6 Sol (Max) | 82.93 |
| 9 | GPT-6 (Non-reasoning) | 82.93 |
| 10 | GPT-5.5 Pro (xHigh) | 78.05 |
| 11 | Claude Opus 5 (Max) | 73.17 |
| 12 | GPT-5.5 (xHigh) | 72.5 |
| 13 | GPT-5.6 Terra (Max) | 70.73 |
| 14 | GPT-5.6 Luna (Max) | 60.98 |
| 15 | GPT-5.4 Pro (xHigh) | 58.54 |
Interactive version: theaggregate.ai/benchmark?slug=frontiermath-tier-4-v2 · How It Works · Data refreshed daily, snapshot 2026-09-05.