LiveMathematicianBench: leaderboard
Continuously updated benchmark testing LLMs on understanding mathematical theorems from newly published arXiv preprints. Multiple-choice questions constructed from theorem statements and proof sketches.
Metric: Accuracy (%). Source: livemathematicianbench.github.io. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.5 Flash | 45.71 | #60 |
| 2 | GPT-5.4 (2026-03-05) (Medium) | 45.24 | #59 (GPT-5.4 (2026-03-05)) |
| 3 | Claude Opus 5 | 42.69 | #12 |
| 4 | Gemini 3.1 Pro (Preview) | 37.59 | #58 |
| 5 | DeepSeek V4 Pro | 30.39 | #102 |
| 6 | DeepSeek V3.2 | 29 | #205 |
| 7 | Mistral Large 3 | 28.31 | #401 |
| 8 | GPT-5 (Medium) | 28.31 | #98 (GPT-5) |
| 9 | DeepSeek R1 | 27.61 | #251 |
| 10 | Gemini 2.5 Pro | 26.91 | #152 |
| 11 | Llama 4 Maverick | 26.45 | #464 |
| 12 | GPT-4.1 | 25.75 | #248 |
| 13 | O4 Mini (Medium) | 24.59 | #180 (O4 Mini) |
| 14 | GPT-5.2 (2025-12-11) (Medium) | 24.59 | |
| 15 | O3 (2025-04-16) (Medium) | 24.36 | #121 (O3 (2025-04-16)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=livemathematicianbench · How It Works · Data refreshed daily, snapshot 2026-10-20.