Vision Arena

View rankings across multimodal, generative AI models capable of understanding and processing visual inputs

Last Updated

Dec 4, 2025

Total Votes

567,510

Total Models

/

Style Control

	Rank Spread (Upper-Lower)
1	1◄─►1	gemini-3-pro	1314	±12	3,603	Google	Proprietary
2	2◄─►5	gemini-2.5-pro	1249	±7	65,142	Google	Proprietary
3	2◄─►7	gpt-5.1-high	1248	±16	1,505	OpenAI	Proprietary
4	2◄─►7	gpt-5.1	1246	±16	1,497	OpenAI	Proprietary
5	2◄─►7	chatgpt-4o-latest-20250326	1237	±7	17,102	OpenAI	Proprietary
6	3◄─►15	gpt-4.5-preview-2025-02-27	1226	±11	2,925	OpenAI	Proprietary
7	3◄─►15	gemini-2.5-flash-preview-09-2025	1226	±11	4,639	Google	Proprietary
8	6◄─►15	gpt-5-chat	1222	±8	38,125	OpenAI	Proprietary
9	6◄─►16	o3-2025-04-16	1218	±7	44,890	OpenAI	Proprietary
10	6◄─►19	gpt-4.1-2025-04-14	1216	±7	40,396	OpenAI	Proprietary
11	6◄─►19	gemini-2.5-flash	1215	±7	39,984	Google	Proprietary
12	6◄─►26	claude-opus-4-20250514-thinking-16k	1210	±15	1,423	Anthropic	Proprietary
13	6◄─►24	gpt-5-high	1209	±8	32,280	OpenAI	Proprietary
14	6◄─►26	qwen3-vl-235b-a22b-instruct	1206	±10	4,841	Alibaba	Apache 2.0
15	6◄─►27	claude-sonnet-4-20250514-thinking-32k	1205	±16	1,313	Anthropic	Proprietary
16	10◄─►26	gpt-4.1-mini-2025-04-14	1202	±8	39,495	OpenAI	Proprietary
17	10◄─►26	o4-mini-2025-04-16	1202	±8	40,020	OpenAI	Proprietary
18	9◄─►31	ernie-5.0-preview-1120	1199	±13	2,282	Baidu	Proprietary
19	10◄─►31	claude-3-7-sonnet-20250219-thinking-32k	1196	±15	1,630	Anthropic	Proprietary
20	12◄─►31	hunyuan-vision-1.5-thinking	1194	±12	2,734	Tencent	Proprietary
21	12◄─►31	o1-2024-12-17	1193	±10	3,694	OpenAI	Proprietary
22	12◄─►31	qwen3-vl-235b-a22b-thinking	1190	±13	2,524	Alibaba	Apache 2.0
23	12◄─►31	claude-opus-4-20250514	1189	±12	2,516	Anthropic	Proprietary
24	13◄─►31	gemini-2.5-flash-lite-preview-06-17-thinking	1188	±8	36,743	Google	Proprietary
25	12◄─►32	claude-sonnet-4-20250514	1187	±13	2,036	Anthropic	Proprietary
26	13◄─►32	qwen-vl-max-2025-08-13	1184	±12	3,313	Alibaba	Proprietary
27	17◄─►32	gpt-5-mini-high	1183	±9	28,912	OpenAI	Proprietary
28	18◄─►32	grok-4-0709	1181	±8	31,159	xAI	Proprietary
29	18◄─►33	gemini-1.5-pro-002	1178	±8	8,916	Google	Proprietary
30	18◄─►36	claude-3-7-sonnet-20250219	1178	±9	4,661	Anthropic	Proprietary
31	18◄─►36	gemini-2.5-flash-lite-preview-09-2025-no-thinking	1177	±10	4,704	Google	Proprietary
32	25◄─►39	gemini-2.0-flash-001	1169	±7	9,738	Google	Proprietary
33	29◄─►43	gpt-4o-2024-05-13	1162	±8	23,280	OpenAI	Proprietary
34	30◄─►43	claude-3-5-sonnet-20241022	1162	±7	10,532	Anthropic	Proprietary
35	30◄─►47	glm-4.5v	1157	±12	3,478	Z.ai	MIT
36	32◄─►47	mistral-medium-2505	1156	±9	11,433	Mistral	Proprietary
37	30◄─►48	step-1o-turbo-202506	1155	±14	2,057	StepFun	Proprietary
38	32◄─►47	gemma-3-27b-it	1154	±9	18,298	Google	Gemma
39	33◄─►48	mistral-medium-2508	1149	±8	33,048	Mistral	Proprietary
40	33◄─►48	claude-3-5-sonnet-20240620	1147	±9	21,632	Anthropic	Proprietary
41	32◄─►53	hunyuan-large-vision	1147	±16	1,466	Tencent	Proprietary
42	33◄─►50	gpt-5-nano-high	1147	±12	4,100	OpenAI	Proprietary
43	33◄─►48	llama-4-maverick-17b-128e-instruct	1146	±9	7,161	Meta	Llama 4
44	35◄─►54	step-3	1141	±12	3,441	StepFun	Apache 2.0
45	35◄─►54	mistral-small-2506	1140	±10	11,644	Mistral	Apache 2.0
46	35◄─►54	gemini-1.5-flash-002	1139	±9	7,256	Google	Proprietary
47	35◄─►56	claude-3-5-haiku-20241022	1135	±15	1,567	Anthropic	Proprietary
48	38◄─►56	gemini-2.0-flash-lite-preview-02-05	1133	±10	4,001	Google	Proprietary
49	43◄─►56	mistral-small-3.1-24b-instruct-2503	1126	±9	27,479	Mistral	Apache 2.0
50	42◄─►56	llama-4-scout-17b-16e-instruct	1126	±10	6,631	Meta	Llama
51	42◄─►56	step-1o-vision-32k-highres	1124	±12	2,844	StepFun	Proprietary
52	43◄─►56	qwen2.5-vl-72b-instruct	1122	±10	3,767	Alibaba	Qwen
53	44◄─►57	gpt-4o-2024-08-06	1119	±12	3,384	OpenAI	Proprietary
54	47◄─►57	gemini-1.5-pro-001	1118	±11	16,735	Google	Proprietary
55	43◄─►59	qwen2.5-vl-32b-instruct	1118	±15	1,490	Alibaba	Apache 2.0
56	47◄─►59	gpt-4-turbo-2024-04-09	1113	±11	13,390	OpenAI	Proprietary
57	55◄─►61	gpt-4o-mini-2024-07-18	1098	±7	17,362	OpenAI	Proprietary
58	55◄─►61	pixtral-large-2411	1094	±9	5,439	Mistral	MRL
59	53◄─►65	gpt-4.1-nano-2025-04-14	1089	±18	1,211	OpenAI	Proprietary
60	57◄─►63	qwen2-vl-72b	1086	±9	5,946	Alibaba	Qwen
61	57◄─►65	qwen-vl-max-1119	1086	±15	1,431	Alibaba	Proprietary
62	59◄─►67	gemini-1.5-flash-8b-001	1070	±10	6,255	Google	Proprietary
63	60◄─►67	claude-3-opus-20240229	1065	±10	15,568	Anthropic	Proprietary
64	59◄─►67	step-1v-32k	1062	±16	1,551	StepFun	Proprietary
65	60◄─►67	gemini-1.5-flash-001	1060	±11	13,261	Google	Proprietary
66	62◄─►70	molmo-72b-0924	1048	±12	3,063	AI2	Apache 2.0
67	62◄─►74	hunyuan-standard-vision-2024-12-31	1044	±21	809	Tencent	Proprietary
68	66◄─►74	llama-3.2-vision-90b-instruct	1033	±8	8,708	Meta	Llama 3.2
69	66◄─►76	qwen2-vl-7b-instruct	1032	±10	5,778	Aliaba	Apache 2.0
70	67◄─►76	pixtral-12b-2409	1025	±9	7,520	Mistral	Apache 2.0
71	66◄─►77	internvl2-26b	1025	±12	5,146	OpenGVLab	MIT
72	67◄─►79	amazon-nova-lite-v1.0	1021	±15	1,856	Amazon	Proprietary
73	67◄─►79	amazon-nova-pro-v1.0	1021	±13	2,338	Amazon	Proprietary
74	67◄─►78	claude-3-sonnet-20240229	1020	±11	12,314	Anthropic	Proprietary
75	69◄─►81	yi-vision	1005	±18	1,220	01 AI	Proprietary
76	71◄─►81	claude-3-haiku-20240307	1003	±12	13,380	Anthropic	Proprietary
77	69◄─►83	c4ai-aya-vision-32b	1001	±22	847	Cohere	CC-BY-NC-4.0
78	72◄─►81	molmo-7b-d-0924	997	±13	2,828	AI2	Apache 2.0
79	75◄─►81	llama-3.2-vision-11b-instruct	991	±11	4,818	Meta	Llama 3.2
80	73◄─►85	nvila-internal-15b-v1	988	±20	1,078	NVIDIA	-
81	75◄─►85	llava-onevision-qwen2-72b-ov	981	±18	1,321	LLaVA	Apache 2.0
82	80◄─►85	llava-v1.6-34b	967	±12	4,531	LLaVA	Apache 2.0
83	79◄─►85	minicpm-v-2_6	965	±15	1,987	OpenBMB	Apache 2.0
84	79◄─►85	cogvlm2-llama3-chat-19b	964	±15	1,991	Zhipu AI	CogVLM2
85	80◄─►85	internvl2-4b	958	±12	3,706	OpenGVLab	MIT
86	86◄─►86	phi-3.5-vision-instruct	922	±15	2,592	Microsoft	MIT
87	87◄─►87	phi-3-vision-128k-instruct	884	±18	1,401	Microsoft	MIT

Vision Arena

Default Leaderboard Plots

Battle Count for Each Combination of Models (without Ties)

Average Win Rate Against All Other Models (Uniform Sampling and No Ties)

Confidence Intervals on Model Strength (via Bootstrapping)

Fraction of Model A Wins for All Non-tied A vs. B Battles