MathBench: leaderboard
Hierarchical math evaluation from arithmetic through college level, testing across 5 progressive difficulty stages with both theoretical and application questions.
Metric: Average Score. Source: github.com. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-05-13) | 70.9 | #325 |
| 2 | Qwen 2 72B Instruct | 67 | #576 |
| 3 | Claude 3 Opus | 63 | #571 |
| 4 | GPT-4 Preview (0125) | 58.8 | #486 |
| 5 | Qwen 1.5 110B Chat | 58.4 | #691 |
| 6 | Llama 3 70B Instruct | 56.4 | #634 |
| 7 | Qwen 2 7B Instruct | 53.4 | #991 |
| 8 | Yi 1.5 34B Chat | 52.2 | #800 |
| 9 | GLM-4 | 51.3 | #648 |
| 10 | Yi-1.5-9B Chat | 51.1 | #1031 |
| 11 | Qwen-72B-Chat | 49.7 | #843 |
| 12 | GLM-4 9B Chat | 47.7 | #916 |
| 13 | internlm2-chat-20B | 42.1 | #999 |
| 14 | GPT-3.5 Turbo (0125) | 41 | #984 |
| 15 | Qwen-14B-Chat | 39.5 | #1169 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mathbench · How It Works · Data refreshed daily, snapshot 2026-10-20.