MathNet - Number Theory: leaderboard

MathNet (2026) is a multilingual set of Olympiad problems with official solutions from national competition booklets of 47 countries; on its problem-solving test set a GPT-5 grader scores each solution from 0 to 7 against the official one, and 6 or more counts as correct. Accuracy on the number theory problems, in percent; higher is better.

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 February 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)82.2#99
2GPT-573.6#178
3Gemini 3 Flash (Preview)73.3#152
4GPT-5 Mini61.5#364
5GPT-5 Nano49.6#906
6DeepSeek V3.2Non-reasoning45.3#419
7Claude Opus 4.6Thinking44.6#113
8Gemini 2.5 Flash42.6#463
9MathNet DeepSeek-R1 (checkpoint unspecified)39.5
10Grok 333#611
11GPT-4.124#505
12Llama 4 Maverick Instruct FP814.4#1023
13GPT-4o7#751
14Ministral 3B2.9#6858

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mathnet-number-theory · How It Works · Data refreshed daily, snapshot 2026-11-21.