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
1	Gemini 4 Carbon (Medium) Google	89.2%
2	Claude Opus 5.5 (Max, Default Fallback) Anthropic	87.7%
3	GPT-6 Astra (Max) OpenAI	86.9%
4	Claude Opus 5.5 (Xhigh, Default Fallback) Anthropic	86.6%
5	GPT-6 Astra (High) OpenAI	86.4%
6	GPT-6 Astra (Xhigh) OpenAI	86.2%
7	GPT-6.1 Sol (Max) OpenAI	86.0%
8	Claude Opus 5.5 (High, Default Fallback) Anthropic	85.8%
9	Claude Opus 5.5 (Medium, Default Fallback) Anthropic	85.7%
10	Gemini 3.8 Flash (High) Google	85.6%
11	GPT-6.1 Sol (Xhigh) OpenAI	85.1%
12	GPT-6 Astra (Medium) OpenAI	85.1%
13	GPT-6.1 Sol (High) OpenAI	84.9%
14	Claude Opus 5.5 (Low, Default Fallback) Anthropic	84.7%
15	GPT-6.1 Terra (Max) OpenAI	84.7%
16	GPT-6 Astra (Low) OpenAI	84.6%
17	Gemini 3.8 Flash (Low) Google	84.5%
18	Gemini 3.8 Flash (Medium) Google	84.2%
19	GPT-6.1 Sol (Medium) OpenAI	83.9%
20	Kimi K3.1 Pro Kimi	83.9%
21	GPT-6.1 Terra (High) OpenAI	83.8%
22	GPT-6.1 Terra (Xhigh) OpenAI	83.8%
23	Muse Wonder (High) Meta	83.8%
24	GPT-6.1 Sol (Low) OpenAI	83.1%
25	GPT-6.1 Terra (Medium) OpenAI	82.8%
26	Qwen3.8 Max (0902) Alibaba	82.8%
27	GPT-6.1 Terra (Low) OpenAI	82.5%
28	Gemini 3.1 Pro Preview Google	82.4%
29	Inkling 2 Small Thinking Machines	81.6%
30	Muse Spark Open (Xhigh) Meta	80.9%
31	DeepSeek V4.1 Flash 1017 (Max) DeepSeek	80.6%
32	Qwen3.7 Plus Alibaba	80.5%
33	DeepSeek V4.1 Flash 1017 (High) DeepSeek	79.7%
34	GPT-6 Luna (Xhigh) OpenAI	79.7%
35	GPT-6 Luna (Max) OpenAI	79.7%
36	Apodex 1.1 Apodex	79.2%
37	MiniMax-M3.1-Flash MiniMax	79.1%
38	Gemini 3.5 Flash-Lite Google	79.0%
39	Ling-3.0-flash-VL InclusionAI	79.0%
40	Qwen3.8-Flash-Next Alibaba	78.6%
41	MiniMax-M3 MiniMax	78.6%
42	GPT-5.3 Codex (Xhigh) OpenAI	78.5%
43	Agnes 2.5 Pro Theta Sapiens AI	78.4%
44	Grok 4.7 (Low) SpaceXAI	78.2%
45	GPT-6 Luna (Medium) OpenAI	77.9%
46	GPT-6 Luna (High) OpenAI	77.5%
47	Qwen3.5 397B A17B (Reasoning) Alibaba	77.3%
48	Step 5 Preview StepFun	76.4%
49	Qwen3.8 27B (Xhigh) Alibaba	76.3%
50	GPT-6.1 Terra (Non-reasoning) OpenAI	76.0%

1
Gemini 4 Carbon (Medium)
Google
89.2%
2
Claude Opus 5.5 (Max, Default Fallback)

データ提供：Artificial Analysis

Command Palette

MMMU-Pro

スコア分布

リーダーボード