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. Status: saturated. 34 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o (2024-05-13) | 70.9 |
| 2 | Qwen 2 72B Instruct | 67 |
| 3 | Claude 3 Opus | 63 |
| 4 | GPT-4 Preview (0125) | 58.8 |
| 5 | Qwen 1.5 110B Chat | 58.4 |
| 6 | Llama 3 70B Instruct | 56.4 |
| 7 | Qwen 2 7B Instruct | 53.4 |
| 8 | Yi 1.5 34B Chat | 52.2 |
| 9 | Yi-1.5-9B Chat | 51.1 |
| 10 | GLM-4 9B Chat | 47.7 |
| 11 | internlm2-chat-20B | 42.1 |
| 12 | GPT-3.5 Turbo (0125) | 41 |
| 13 | Qwen-14B-Chat | 39.5 |
| 14 | Llama 3 8B Instruct | 36.7 |
| 15 | internlm2-chat-7B | 34.1 |
Interactive version: theaggregate.ai/benchmark?slug=mathbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.