MMMU-Pro authors

Massive Multi-discipline Multimodal Understanding Pro

A harder multimodal benchmark for frontier models that combines text with images, diagrams, charts, and academic visual reasoning tasks.

In index
Multimodal
Source ↗

Models41

Top result83.9%Gemini 3.1 Pro

Top-3 spread0.9 pts

Year2024

Result and price

Results

41 results

#ModelResult

01

Gemini 3.1 Pro83.9%

02

Gemini 3.5 Flash83.6%

03

GPT-5.6 Sol83.0%

04

Qwen3.8 Max82.3%

05MAKimi K381.6%

05

Seed 2.1 Pro81.6%

07

GPT-5.581.2%

07

GPT-5.481.2%

09

Gemini 3 Pro81.0%

10

GPT-5.6 Terra80.7%

11

Muse Spark80.4%

12

Seed 2.1 Turbo80.1%

13

GPT-5.279.5%

14MAKimi K2.679.4%

15DSdots3-note Preview79.1%

16

Qwen3.7 Plus79.0%

16

Qwen3.5 397B79.0%

18

Qwen3.6 Plus78.8%

19MAKimi K2.5 (Reasoning)78.5%

19MAKimi K2.578.5%

21

GPT-5.6 Luna78.4%

22

MiniMax M378.1%

22

Grok 4.378.1%

24

MiMo-V2.577.9%

25

Claude Opus 4.677.3%

26

Gemma 4 31B76.9%

27

GPT-5.4 mini76.6%

28STStep 5 Preview76.0%

29

Qwen3.6-27B75.8%

30

Qwen3.6-35B-A3B75.3%

31

Grok 4.2075.2%

32TMInkling-Small74.0%

32

Muse Glimmer 30B74.0%

34

Gemma 4 26B A4B73.8%

35TMInkling73.5%

36INInterfaze Beta71.1%

37

Claude Opus 4.570.6%

38

Gemma 4 12B69.1%

39

GPT-5.4 nano66.1%

40COCommand A+63.0%

41LILFM2.5-VL-3B30.5%