Benchmarks / Other

Arena Vision

Human preference on prompts with image inputs (multimodal understanding); Bradley-Terry rating.

107models
–Open Index weight
–last updated
LMArenapublisher

Arena Vision

Higher is better · best effort per model

Chart settings
107 of 107 selected

full table below

#ModelEffortScoreCost / taskHarness
1Fable 5high1308––
2Qwen 3.8 Maxmax1301––
3Opus 4.6high1299––
4Opus 4.7max1298––
5Gemini 3.7 Flashhigh1297––
6GPT-6.1 Solmax1294––
7Muse Spark 1.2xhigh1293––
8Opus 5.5high1293––
9Muse Spark 1.3max1290––
10Gemini 3 Prohigh1289––
11Gemini 3.8 Flashhigh1289––
12Opus 5high1287––
13GPT-5.6 Solxhigh1287––
14Fable 5.1max1286––
15GPT-6 Astramax1286––
16GPT-5.5xhigh1286––
17Opus 4.8high1286––
18Gemini 3.5 Flashmedium1285––
19GPT-5.4high1285––
20Gemini 3.6 Flashhigh1281––
21Grok 4.5max1280––
22Muse Spark 1.1max1280––
23Gemini 3.1 Prohigh1280––
24GPT-5.2xhigh1278––
25GPT-5.5 Instant–1276––
26Sonnet 4.6max1275––
27GLM 5.3 Flashmax1273––
28Gemini 3 Flashhigh1273––
29GPT-6 Solmax1270––
30GPT-5.6 Terraxhigh1268––
31Sonnet 5.5xhigh1266––
32Gemini 3.5 Flash-Litexhigh1266––
33Kimi K2.6max1265––
34Qwen 3.7 Plushigh1264––
35Sonnet 5high1263––
36Gemma 4 31Bmax1262––
37Grok 4.6high1262––
38DeepSeek V4.1 Flashmax1260––
39GPT-5.6 Lunaxhigh1258––
40Grok 4.20xhigh1256––
41Kimi K2.5max1252––
42GPT-5.1high1251––
43GPT-5.4 Minihigh1251––
44MiMo V2.6 Flashhigh1248––
45Gemini 2.5 Prohigh1248––
46Grok 4.7xhigh1248––
47MiMo V2.6 Prohigh1246––
48Qwen 3.5 397B-A17Bmax1246––
49Qwen 3.8 27Bmax1244––
50Grok 4.3xhigh1243––
51Gemma 4 26Bxhigh1242––
52GPT-6 Lunamax1241––
53MiniMax M3max1236––
54Gemini 3.1 Flash-Litehigh1235––
55MiMo V2.5xhigh1234––
56GLM 5 Vhigh1231––
57Gemini 2.5 Flashmedium1229––
58Qwen 3.5 122B-A10Bnone1226––
59GPT-5high1226––
60GPT-4.5none1225––
61Qwen 3.5 27Bhigh1217––
62MiMo V2–1216––
63o3high1216––
64Qwen 3high1215––
65GPT-4.1high1214––
66Inkling Smallxhigh1208––
67Opus 4medium1208––
68Sonnet 4high1206––
69GPT-4.1 Minihigh1201––
70o4 minihigh1201––
71Mistral Medium 3.5high1200––
72Mistral Large 3none1200––
73GPT-5.4 Nanohigh1199––
74Grok 4.1 Fasthigh1196––
75Sonnet 3.7medium1193––
76o1high1193––
77Gemini 2.5 Flash Litenone1188––
78Grok 4.0709–1184––
79GPT-5 Minihigh1183––
80Gemini 1.5 Pronone1179––
81Gemini 2.0 Flashmedium1172––
82GPT-4ohigh1162––
83GLM 4.6 Vlow1161––
84Sonnet 3.5medium1160––
85Gemma 3 27B ITmedium1160––
86Mistral Medium 2508–1159––
87Mistral Medium 2505–1155––
88GLM 4.5 Vnone1153––
89GPT-5 Nanohigh1146––
90Llama 4 Maverickmedium1145––
91Gemini 1.5 Flashnone1141––
92Mistral Small 2506–1140––
93Gemini 2.0 Flash-Litenone1136––
94Llama 4 Scoutmedium1127––
95Mistral Small 3.1none1127––
96Haiku 3.5medium1127––
97Qwen 2.5–1121––
98GPT-4high1112––
99GPT-4o minihigh1097––
100GPT-4.1 Nanohigh1089––
101Qwen 2–1084––
102Opus 3none1061––
103Llama 3.2–1032––
104Sonnet 3none1016––
105Haiku 3medium1000––
106Qwen 2 72Bnone978––
107Llama 3 19b–963––