MMMU-Pro (official)

Official MMMU-Pro leaderboard. Shown for reference.

Published by

MMMU

Category

Multimodal

Index weight

Reference only

Models

25

Data as of

9 Sept 2026

MMMU (Yue et al.).

1. 1Gemini 3 Pro81.0%
2. 2Gemini 3.1 Pro80.5%
3. 3Muse Spark80.4%
4. 4GPT-5.179.0%
5. 5Gemma-4-31B-IT76.9%
6. 6o376.4%
7. 7GPT-5.176.0%
8. 8Claude Opus 4.573.9%
9. 9Gemma 4 26B A4B73.8%
10. 10Claude Sonnet 4.568.9%
11. 11Qwen3-VL 235B-A22B68.1%
12. 12Gemini 2.5 Pro68.0%
13. 13Seed 1.666.4%
14. 14Gemma 4 E4B52.6%
15. 15GPT-4o51.9%

SearchAll providersAlibabaAnthropicByteDanceGoogleMetaMicrosoftNVIDIAOpenAIOpenBMB / LLaVAOpen weights25 of 25

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Gemini 3 ProGoogle	81.0%	Gemini 3.0 Pro	61.6	–
2	Gemini 3.1 ProGoogle	80.5%	–	60.7	–
3	Muse SparkMeta	80.4%	–	–	–
4	GPT-5.1OpenAI	79.0%	–	55.2	–
5	Gemma-4-31B-ITNVIDIAopen	76.9%	Gemma 4 31B	55.4	–
6	o3OpenAI	76.4%	o3	60.3	–
7	GPT-5.1OpenAI	76.0%	GPT-5.1	59.4	–
8	Claude Opus 4.5Anthropic	73.9%	Claude Opus 4.5	58.0	–
9	Gemma 4 26B A4BGoogleopen	73.8%	Gemma 4 26B A4B	54.0	–
10	Claude Sonnet 4.5Anthropic	68.9%	Claude Sonnet 4.5	52.3	–
11	Qwen3-VL 235B-A22BAlibabaopen	68.1%	–	49.6	–
12	Gemini 2.5 ProGoogle	68.0%	–	54.1	–
13	Seed 1.6ByteDance	66.4%	–	–	–
14	Gemma 4 E4BGoogleopen	52.6%	Gemma 4 E4B	45.4	–
15	GPT-4oOpenAI	51.9%	0513	42.3	–
16	Claude 3.5 SonnetAnthropic	51.5%	Claude 3.5 Sonnet	45.1	–
17	Gemini 1.5 ProGoogle	46.9%	0801	42.9	–
18	Qwen2 Vl 72BAlibaba	46.2%	–	–	–
19	Qwen2.5-VL 72B InstructAlibabaopen	46.2%	–	–	–
20	Gemma 4 E2BGoogleopen	44.2%	Gemma 4 E2B	42.2	–
21	Llama 3.2 90BMeta	39.5%	Llama 3.2 90B	–	–
22	Qwen2.5-VL 7B InstructAlibabaopen	38.3%	–	–	–
23	GPT-4o miniOpenAI	37.6%	GPT-4o mini	35.8	–
24	OMinicpm V 2.6OpenBMB / LLaVA	27.2%	MiniCPM-V 2.6	–	–
25	Phi 3.5 VisionMicrosoft	19.7%	–	–	–