Benchmarks / Reasoning & Math
FrontierMath
Research-level mathematics problems maintained by Epoch AI.
maintained by Epoch AI · 173 observations on record
Versions
2 versions
| Version | Tasks |
|---|---|
| FrontierMath current frontiermath | — |
| FrontierMath Tier 4 frontiermath-tier-4 | — |
Current leaderboard
FrontierMath · Accuracy · top 25 of 101 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | GPT 5.5 Pro Pre Releasehigh effort OpenAI | 52.4% |
| 2 | GPT 5.5 Pre Releasexhigh effort OpenAI | 51.7% |
| 3 | GPT 5.5 Pro Pre Releasexhigh effort OpenAI | 51.0% |
| 4 | GPT 5.4 Pro (2026-03-05)xhigh effort OpenAI | 50.0% |
| 5 | GPT 5.4 (2026-03-05)xhigh effort OpenAI | 47.6% |
| 6 | Claude Opus 4.8max effort Anthropic | 47.2% |
| 7 | Claude Opus 4.7xhigh effort Anthropic | 43.8% |
| 8 | Claude Opus 4.6max effort Anthropic | 40.7% |
| 9 | GPT 5.2 (2025-12-11)xhigh effort OpenAI | 40.7% |
| 10 | GPT 5.2 (2025-12-11)high effort OpenAI | 40.3% |
| 11 | Claude Opus 4.632K budget Anthropic | 40.0% |
| 12 | Claude Opus 4.664K budget Anthropic | 39.7% |
| 13 | Muse Spark Meta AI | 39.0% |
| 14 | Gemini 3.5 Flashhigh effort Google DeepMind | 39.0% |
| 15 | Kimi K2 6 Moonshot AI | 39.0% |
| 16 | Claude Opus 4.6 Anthropic | 38.3% |
| 17 | Gemini 3 Pro Preview Google DeepMind | 37.6% |
| 18 | GPT 5.2 (2025-12-11)medium effort OpenAI | 36.9% |
| 19 | Gemini 3.1 Pro Preview Google DeepMind | 36.9% |
| 20 | Gemini 3 Flash Preview Google DeepMind | 35.6% |
| 21 | GLM 5.1 Z.ai | 33.5% |
| 22 | GPT 5 (2025-08-07)high effort OpenAI | 32.4% |
| 23 | Claude Sonnet 4.616K budget Anthropic | 32.4% |
| 24 | GPT 5.1 (2025-11-13)high effort OpenAI | 31.0% |
| 25 | Gemini 2.5 Deep Think 2025 08.01 Webapp Google,Google DeepMind | 29.0% |
One row per evaluated system — reasoning-effort variants rank separately.