MMMU-Pro

MMMU-Pro: multimodal college-level understanding; accuracy.

160models

–Open Index weight

–last updated

Artificial Analysispublisher

#	Model	Effort	Score	Cost / task	Harness
1	Opus 5.5	max	87.7%	–	–
2	GPT-6 Astra	max	86.9%	–	–
3	GPT-6.1 Sol	max	86.0%	–	–
4	Gemini 3.8 Flash	high	85.6%	–	–
5	Gemini 3.7 Flash	high	85.5%	–	–
6	Opus 5	max	84.7%	–	–
7	Gemini 3.5 Flash	high	84.3%	–	–
8	GPT-5.6 Sol	max	83.4%	–	–
9	Gemini 3.6 Flash	high	83.2%	–	–
10	GPT-6 Sol	max	82.9%	–	–
11	Qwen 3.8 Max	max	82.8%	–	–
12	Gemini 3.1 Pro	–	82.4%	–	–
13	Muse Spark 1.3	xhigh	82.0%	–	–
14	GPT-5.5	medium	81.2%	–	–
15	Muse Spark Open (Xhigh)	xhigh	80.9%	–	–
16	GPT-5.6 Terra	max	80.7%	–	–
17	Kimi K3	max	80.5%	–	–
18	Muse Spark	–	80.5%	–	–
19	Qwen 3.7 Plus	–	80.5%	–	–
20	Grok 4.5	high	80.4%	–	–
21	Gemini 3 Pro	high	80.2%	–	–
22	Gemini 3 Flash	–	79.9%	–	–
23	Qwen 3.8 Flash	–	79.8%	–	–
24	GPT-6 Luna	xhigh	79.7%	–	–
25	Kimi K2.6	–	79.4%	–	–
26	Apodex 1.1	–	79.2%	–	–
27	Gemini 3.5 Flash-Lite	high	79.0%	–	–
28	Ling 3.0 Flash	–	79.0%	–	–
29	Opus 4.7	max	78.8%	–	–
30	GPT-5.6 Luna	max	78.6%	–	–
31	MiniMax M3	–	78.6%	–	–
32	GPT-5.3 Codex	xhigh	78.5%	–	–
33	GPT-5.4	xhigh	78.4%	–	–
34	Grok 4.7	low	78.2%	–	–
35	Grok 4.3	high	78.1%	–	–
36	Qwen 3.6 Plus	–	78.0%	–	–
37	Sonnet 5	max	77.3%	–	–
38	Qwen 3.5 397B-A17B	–	77.3%	–	–
39	DeepSeek V4.1 Flash	max	77.0%	–	–
40	Grok Build 0.1	–	76.5%	–	–
41	Mistral Large 4	–	76.4%	–	–
42	Step 5 Preview	–	76.4%	–	–
43	Qwen 3.8 27B	xhigh	76.3%	–	–
44	GPT-5.2 Codex	xhigh	76.3%	–	–
45	Gemini 3.1 Flash-Lite	high	75.5%	–	–
46	GPT-5.1	high	75.5%	–	–
47	Agnes 2.5 Pro Beta	–	75.4%	–	–
48	Opus 4.6	max	75.4%	–	–
49	MiMo V2.5	–	75.4%	–	–
50	Kimi K2.5	–	75.4%	–	–
51	Step 3.7 Flash	high	75.3%	–	–
52	Qwen 3.5 27B	–	75.0%	–	–
53	Qwen 3.6 35B-A3B	–	75.0%	–	–
54	Qwen 3.5 122B-A10B	–	75.0%	–	–
55	Gemini 2.5 Pro	–	74.9%	–	–
56	DeepSeek V4 Flash	max	74.8%	–	–
57	Qwen 3.6 27B	–	74.6%	–	–
58	GPT-5.2	medium	74.6%	–	–
59	Grok 4.20	–	74.6%	–	–
60	GPT-5	medium	74.3%	–	–
61	Muse Glimmer (High)	high	74.3%	–	–
62	Opus 4.5	–	74.0%	–	–
63	Inkling Small	–	74.0%	–	–
64	MiMo V2	–	73.9%	–	–
65	GPT-5 Codex	high	73.8%	–	–
66	Inkling	xhigh	73.5%	–	–
67	Gemma 4 31B	–	73.4%	–	–
68	Sonnet 4.6	max	73.3%	–	–
69	GPT-5.4 Mini	xhigh	73.3%	–	–
70	Gemini 2.5 Flash	–	73.1%	–	–
71	MiMo V2.6 Flash	–	73.1%	–	–
72	GLM 5 V	–	72.8%	–	–
73	Qwen 3.5 35B-A3B	–	72.7%	–	–
74	GPT-5.1 Codex	high	72.5%	–	–
75	Qwen 3.5	none	70.5%	–	–
76	GPT-5 Mini	high	70.1%	–	–
77	o3	–	70.1%	–	–
78	Gemma 4 12B	–	69.7%	–	–
79	Gemma 4 26B	–	69.2%	–	–
80	o4 mini	high	69.2%	–	–
81	Qwen 3.5 9B	–	69.2%	–	–
82	GPT-5.1 Codex Mini	high	69.0%	–	–
83	Grok 4	–	68.8%	–	–
84	Sonnet 4.5	–	68.7%	–	–
85	Qwen 3	–	68.7%	–	–
86	Doubao Seed Code	–	68.1%	–	–
87	Opus 4.1	–	67.9%	–	–
88	EXAONE 4.5 33B (Reasoning)	–	67.3%	–	–
89	DiffusionGemma 26B A4B	–	66.5%	–	–
90	GPT-5.4 Nano	xhigh	65.4%	–	–
91	Qwen 3.5 4B	–	65.4%	–	–
92	Gemini 2.5 Flash Lite	–	65.0%	–	–
93	Mistral Medium 3.5	high	64.9%	–	–
94	ERNIE 5.0 Thinking Preview	–	64.6%	–	–
95	Nova 2.0 Pro Preview (Medium)	medium	64.5%	–	–
96	JT-4.1 Flash 236B A21B (Non-reasoning)	none	64.1%	–	–
97	Step3 VL 10B	–	64.0%	–	–
98	Nova 2.0 Lite (High)	high	63.8%	–	–
99	Grok 4.1 Fast	–	63.3%	–	–
100	Command A	–	63.2%	–	–
101	Nova 2.0 Pro Preview (Low)	low	62.7%	–	–
102	Nova 2.0 Lite (Medium)	medium	62.5%	–	–
103	Sonnet 4	none	62.4%	–	–
104	Llama 4 Maverick	none	62.1%	–	–
105	Nova 2.0 Omni (Medium)	medium	61.9%	–	–
106	Grok 4 Fast	–	61.8%	–	–
107	GPT-4.1	none	61.2%	–	–
108	GPT-5 Nano	high	61.0%	–	–
109	Sonnet 3.7	none	60.1%	–	–
110	Nova 2.0 Omni (Low)	low	59.8%	–	–
111	Magistral Medium 1.2	medium	59.7%	–	–
112	GPT-4.1 Mini	none	58.7%	–	–
113	Haiku 4.5	–	58.6%	–	–
114	Nova 2.0 Lite (Low)	low	58.0%	–	–
115	Apriel-v1.5-15B-Thinker	–	57.1%	–	–
116	Mistral Small 4	–	56.8%	–	–
117	GPT-4o	none	56.3%	–	–
118	Mistral Large 3	none	55.7%	–	–
119	Magistral Small 1.2	–	55.5%	–	–
120	Gemini 1.5 Pro	none	55.0%	–	–
121	Mistral Medium 3.1	none	54.2%	–	–
122	Nemotron 3 Nano Omni 30B A3B Reasoning	–	53.2%	–	–
123	Mistral Medium 3	none	53.0%	–	–
124	Llama 4 Scout	none	52.9%	–	–
125	NVIDIA Nemotron Nano 12B v2 VL (Reasoning)	–	52.9%	–	–
126	Gemma 4 E4B (Reasoning)	–	51.4%	–	–
127	Gemma 4 E4B (Non-reasoning)	none	51.2%	–	–
128	Pixtral Large	none	50.6%	–	–
129	GLM 4.5 V	–	50.5%	–	–
130	Nova 2.0 Omni (Non-reasoning)	none	49.9%	–	–
131	Ministral 3 14B	none	49.8%	–	–
132	Nova 2.0 Lite (Non-reasoning)	none	49.0%	–	–
133	GLM 4.6 V	–	48.6%	–	–
134	Gemini 1.5 Flash	none	48.4%	–	–
135	Gemma 3 27B	none	48.0%	–	–
136	Mistral Small 3.2	none	48.0%	–	–
137	Ministral 3 8B	none	46.0%	–	–
138	Haiku 3.5	none	45.6%	–	–
139	Devstral Small 2	none	44.6%	–	–
140	Gemma 4 E2B (Reasoning)	–	44.6%	–	–
141	NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)	none	44.5%	–	–
142	Nova Pro	none	44.3%	–	–
143	Qwen 3.5 2B	–	43.0%	–	–
144	Gemma 4 E2B (Non-reasoning)	none	41.8%	–	–
145	GPT-4o mini	none	41.5%	–	–
146	GPT-4.1 Nano	none	40.1%	–	–
147	Llama 3.2 90b	none	39.5%	–	–
148	Ministral 3 3B	none	38.1%	–	–
149	MiniCPM-V 4.6 1.3B	none	37.9%	–	–
150	Nova Lite	none	37.8%	–	–
151	Gemma 3 12B	none	37.5%	–	–
152	Molmo2-8B	none	37.5%	–	–
153	Haiku 3	none	30.8%	–	–
154	Gemma 3 4B	none	29.9%	–	–
155	Llama 3.2 11b	none	29.3%	–	–
156	LFM2.5-VL-1.6B	none	26.5%	–	–
157	Gemma 3n E4B Instruct	none	26.2%	–	–
158	Qwen 3.5 0.8B	–	25.8%	–	–
159	Molmo 7B-D	none	24.5%	–	–
160	Phi-4 Multimodal Instruct	none	14.5%	–	–