MMMUMultimodal

MMMU (Massive Multi-discipline Multimodal Understanding) was released by multiple institutions in 2024. It contains about 11,500 university-level multimodal questions across 30 subjects in art, business, science, medicine, humanities, and engineering, requiring joint understanding of images and text.

Ranking Sources

#ModelScore

1

Gemini 3 ProGoogle

87.2%

2

GPT-5.2OpenAI

86.7%

3

Qwen3.6-PlusAlibaba

86.0%

4

GPT-5.1OpenAI

85.4%

5

Qwen3.5-397B-A17BAlibaba

85.0%

6

Kimi-K2.5Moonshot AI

84.3%

7

GPT-5OpenAI

84.2%

8

Qwen3.6-27BAlibaba

82.9%

9

o3OpenAI

82.9%

10

Qwen3.6-35B-A3BAlibaba

81.7%

25 models total

Back to benchmarks