MathVision — leaderboard

Metric: Overall Accuracy (%). Source: mathllm.github.io. 164 models tracked.

Top models

#ModelScore
1GPT-5.496.1
2Kimi K2.693.2
3Gemini 3 Flash90.6
4Gemini 3.1 Pro (Preview) (Thinking)89.8
5Qwen 3.5 397B A17B88.6
6Gemini 3 Pro86.6
7Qwen 3.5 122B A10B86.2
8Qwen 3.5 27B86
9Gemma 4 31B85.6
10Kimi K2.5 (Thinking)85
11Kimi K2.584.2
12Qwen 3.5 35B A3B83.9
13GPT-5.283
14Gemma 4 26B A4B82.4
15Human Expert82

Interactive version: theaggregate.ai/benchmark?slug=mathvision · How the rankings work · Data refreshed daily, snapshot 2026-07-22.