MMMU-Pro

MMMUの拡張ベンチマーク。近道や推測の戦略を排除し、30の学問分野にわたってマルチモーダルモデルをより厳密にテストする。

モデル: 126 現在、 303 合計
中央値: 70.8%
範囲: 14.5%–89.2%

スコア分布

最低: 14.5%
中央値: 70.8%
最高: 89.2%

リーダーボード

非推奨のモデルを含める

ランク	モデル	スコア
51	GPT-6 Luna (Low) OpenAI	76.0%
52	Agnes 2.5 Pro Beta Sapiens AI	75.4%
53	Qwen3.6 35B A3B (Reasoning) Alibaba	75.0%
54	Qwen3.5 122B A10B (Reasoning) Alibaba	75.0%
55	DeepSeek V4 Flash Vision (Max) DeepSeek	74.8%
56	Muse Glimmer (High) Meta	74.3%
57	Qwen3.8 27B (Medium) Alibaba	74.2%
58	Inkling Small Thinking Machines	74.0%
59	Qwen3.8 27B (Low) Alibaba	73.8%
60	Inkling (Xhigh) Thinking Machines	73.5%
61	Gemma 4 31B (Reasoning) Google	73.4%
62	MiMo-V2.6-Flash Xiaomi	73.1%
63	Qwen3.6 35B A3B (Non-reasoning) Alibaba	71.0%
64	Qwen3.5 Omni Plus Alibaba	70.5%
65	Gemma 4 31B (Non-reasoning) Google	70.3%
66	Qwen3.5 122B A10B (Non-reasoning) Alibaba	70.3%
67	o3 OpenAI	70.1%
68	Qwen3.8 27B (Non-reasoning) Alibaba	69.9%
69	Gemma 4 12B (Reasoning) Google	69.7%
70	Gemma 4 26B A4B (Reasoning) Google	69.2%
71	Qwen3.5 9B (Reasoning) Alibaba	69.2%
72	DeepSeek V4.1 Flash 1017 (Non-reasoning) DeepSeek	69.2%
73	Doubao Seed Code ByteDance Seed	68.1%
74	EXAONE 4.5 33B (Reasoning) LG AI Research	67.3%
75	Qwen3.5 9B (Non-reasoning) Alibaba	66.8%
76	Gemma 4 26B A4B (Non-reasoning) Google	66.7%
77	DiffusionGemma 26B A4B Google	66.5%
78	Qwen3.5 4B (Reasoning) Alibaba	65.4%
79	Mistral Medium 3.5 Mistral	64.9%
80	Qwen3.5 Omni Flash Alibaba	64.7%
81	ERNIE 5.0 Thinking Preview Baidu	64.6%
82	Nova 2.0 Pro Preview (Medium) Amazon	64.5%
83	JT-4.1 Flash 236B A21B (Non-reasoning) China Mobile	64.1%
84	Step3 VL 10B StepFun	64.0%
85	Nova 2.0 Lite (High) Amazon	63.8%
86	Command A+ Cohere	63.2%
87	Nova 2.0 Pro Preview (Low) Amazon	62.7%
88	Nova 2.0 Lite (Medium) Amazon	62.5%
89	Llama 4 Maverick Meta	62.1%
90	GPT-6 Luna (Non-reasoning) OpenAI	62.1%
91	Qwen3.5 4B (Non-reasoning) Alibaba	62.1%
92	Gemma 4 12B (Non-reasoning) Google	62.0%
93	Nova 2.0 Omni (Medium) Amazon	61.9%
94	Qwen3 Omni 30B A3B (Reasoning) Alibaba	60.2%
95	Nova 2.0 Omni (Low) Amazon	59.8%
96	Magistral Medium 1.2 Mistral	59.7%
97	Nova 2.0 Lite (Low) Amazon	58.0%
98	Mistral Small 4 (Reasoning) Mistral	56.8%
99	Magistral Small 1.2 Mistral	55.5%
100	Qwen3 Omni 30B A3B Instruct Alibaba	55.5%

51
GPT-6 Luna (Low)
OpenAI
76.0%
52
Agnes 2.5 Pro Beta

データ提供：Artificial Analysis

Command Palette

MMMU-Pro

スコア分布

リーダーボード