U-MATH — leaderboard
Metric: Accuracy (%). Source: huggingface.co. 28 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 86.82 |
| 2 | Gemini 2.0 Flash (01-21) (Thinking) | 83.64 |
| 3 | O3 Mini | 82.18 |
| 4 | DeepSeek R1 | 80.73 |
| 5 | O1 Mini | 76.27 |
| 6 | QwQ 32B-Preview | 73.09 |
| 7 | QVQ-72B-Preview | 65 |
| 8 | DeepSeek V3 | 62.64 |
| 9 | Gemini 1.5 Pro | 60.09 |
| 10 | Qwen2.5-Math-72B-Instruct | 59.45 |
| 11 | Gemini 1.5 Flash | 57.82 |
| 12 | Athene-V2-Chat | 54.91 |
| 13 | Phi-4 | 54.45 |
| 14 | Qwen 2.5 32B Instruct | 52.36 |
| 15 | Qwen 2.5 72B Instruct | 51.18 |
Interactive version: aibenchmarks.dev/benchmark?slug=u-math · How the rankings work · Data refreshed daily, snapshot 2026-07-08.