MathNet - Number Theory: leaderboard
Metric: Problem-solving accuracy (%) on the number theory problems on MathNet-Solve-Test, Olympiad problems from official national competition booklets of 47 countries; a GPT-5 grader scores each solution 0-7 against the official solution and 6 or more counts as correct; image-capable models get the figures, text-only models a text description of them; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.2 |
| 2 | GPT-5 | 73.6 |
| 3 | Gemini 3 Flash (Preview) | 73.3 |
| 4 | GPT-5 Mini | 61.5 |
| 5 | GPT-5 Nano | 49.6 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 45.3 |
| 7 | Claude Opus 4.6 (Thinking) | 44.6 |
| 8 | Gemini 2.5 Flash | 42.6 |
| 9 | Grok 3 | 33 |
| 10 | GPT-4.1 | 24 |
| 11 | Llama 4 Maverick Instruct FP8 | 14.4 |
| 12 | GPT-4o | 7 |
| 13 | Ministral 3B | 2.9 |
Interactive version: theaggregate.ai/benchmark?slug=mathnet-number-theory · How It Works · Data refreshed daily, snapshot 2026-10-07.