U-MATH: leaderboard
U-MATH (Toloka) — university-level mathematics problems spanning algebra through multivariable calculus; accuracy (%).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 | 86.82 | #183 |
| 2 | Gemini 2.0 Flash (01-21) (Thinking) | 83.64 | #322 (Gemini 2.0 Flash (01-21)) |
| 3 | O3 Mini | 82.18 | #272 |
| 4 | DeepSeek R1 | 80.73 | #251 |
| 5 | O1 Mini | 76.27 | #323 |
| 6 | QwQ 32B-Preview | 73.09 | #651 |
| 7 | QVQ-72B-Preview | 65 | #440 |
| 8 | DeepSeek V3 | 62.64 | #321 |
| 9 | Gemini 1.5 Pro | 60.09 | #445 |
| 10 | Qwen2.5-Math-72B-Instruct | 59.45 | #717 |
| 11 | Gemini 1.5 Flash | 57.82 | #618 |
| 12 | Phi-4 | 54.45 | #706 |
| 13 | Qwen 2.5 32B Instruct | 52.36 | #498 |
| 14 | Qwen 2.5 72B Instruct | 51.18 | #454 |
| 15 | GPT-4o (2024-08-06) | 50.18 | #336 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=u-math · How It Works · Data refreshed daily, snapshot 2026-10-20.