MathVision

A visual mathematics benchmark that tests whether a model can solve math problems grounded in diagrams, equations, figures, and other visual inputs.

20 models

Higher is better.

1Qwen3.8 MaxAlibaba95.2
2Kimi K3Moonshot AI94.3
3Seed 2.1 ProByteDance92.6
4Qwen3.8-Omni-FlashAlibaba91.8
5Qwen3.8-Flash-NextAlibaba90.6
6Qwen3.7 PlusAlibaba90.3
7Qwen3.8-27BAlibaba90.0
8Qwen3.5 397BAlibaba88.6
9Qwen3.6 PlusAlibaba88.0
10dots3-note PreviewDots Studio87.7
11Kimi K2.6Moonshot AI87.4
12Gemini 3 ProGoogle86.6
13Qwen3.5-122B-A10BAlibaba86.2
14Qwen3.5-27BAlibaba86.0
15Kimi K2.5Moonshot AI84.2
16Qwen3.5-35B-A3BAlibaba83.9
17GPT-5.2OpenAI83.0
18Gemma 4 12BGoogle79.7
19Nemotron 3.5 Super VL 120B A12BNVIDIA78.2
20Claude Opus 4.5Anthropic74.3

Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric: Image + math reasoning
Tasks: Visually grounded math problems
Authors: Qwen (2026)
Source: Qwen3.6 launch benchmarks ↗

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.