多模態大模型排行榜：視覺大模型排名

依真人匿名對比投票的 Arena 得分排名（LMArena 視覺競技場）：使用者上傳圖片並提問，兩個匿名模型作答後投票。

綜合 OCR 文字辨識 圖表理解 作業題 圖片描述 物體／人物辨識 看圖寫作 幽默理解 英文 中文
綜合第一資料 2026年10月19日
Claude Fable 5 (High)Anthropic · 1325

依 LMArena 視覺競技場 2026年10月19日 發布的資料，Anthropic 的 Claude Fable 5 (High) 以 1325 分（14,714 票）在 167 個模型中排第一。

完整榜單 · 綜合

共 167 個
全部 開放權重

預設依 LMArena 得分降序；± 為 95% 信賴區間半寬；分項欄為該模型在對應分榜的得分，空白為 –。

# 模型 LMArena · 綜合 分項得分 授權
得分 名次 票數 OCR 文字辨識 圖表理解 作業題 圖片描述
1 MPhi-3 Vision 128K InstructMicrosoft開放權重 812±19 167 1,401 – – – – MIT
2 MPhi-3.5 Vision InstructMicrosoft開放權重 851±16 166 2,592 – – – – MIT
3 OMiniCPM-V 2.6OpenBMB開放權重 912±16 165 1,987 – – – – Apache 2.0
4 LLLaVA-OneVision Qwen2 72B OV開放權重 923±18 164 1,321 – – – – Apache 2.0
5 CogVLM2 Llama3 Chat 19BZ.ai開放權重 928±15 163 1,991 – – – – CogVLM2
6 LLLaVA 1.6 34B開放權重 936±12 162 4,531 – – – – Apache 2.0
7 Nnvila-internal-15b-v1NVIDIA開放權重 940±20 161 1,077 – – – – -
8 OInternVL2 4BOpenGVLab開放權重 941±13 160 3,703 – – – – MIT
9 Claude 3 Haiku (20240307)Anthropic 950±13 159 13,380 – – – – Proprietary
10 Molmo 7B-D (0924)Ai2開放權重 959±14 158 2,815 – – – – Apache 2.0
11 Llama 3.2 11B Vision InstructMeta開放權重 966±11 157 4,817 – – – – Llama 3.2
12 0Yi Vision01.AI 978±18 156 1,219 – – – – Proprietary
13 AAmazon Nova Pro 1.0Amazon 980±14 155 2,335 – – – – Proprietary
14 Claude 3 Sonnet (20240229)Anthropic 984±11 154 12,314 – – – – Proprietary
15 AAmazon Nova Lite 1.0Amazon 990±15 152 1,854 – – – – Proprietary
16 Qwen2-VL-7B-InstructAlibaba開放權重 990±11 153 5,766 – – – – Apache 2.0
17 Aya Vision 32BCohere開放權重 996±22 151 847 – – – – CC-BY-NC-4.0
18 Hunyuan Standard Vision (2024-12-31)Tencent 999±21 150 809 – – – – Proprietary
19 Llama 3.2 90B Vision InstructMeta開放權重 1000±9 149 8,682 – – – – Llama 3.2
20 OInternVL2 26BOpenGVLab開放權重 1005±13 148 5,148 – – – – MIT
21 Gemini 1.5 Flash (001)Google 1008±12 147 13,260 – – – – Proprietary
22 Pixtral 12B (2409)Mistral AI開放權重 1009±10 146 7,511 – – – – Apache 2.0
23 Molmo 72B (0924)Ai2開放權重 1012±13 145 3,048 – – – – Apache 2.0
24 Claude 3 Opus (20240229)Anthropic 1023±10 144 15,565 – – – – Proprietary
25 Gemini 1.5 Flash-8B (001)Google 1044±11 143 6,243 – – – – Proprietary
26 SStep-1V 32KStepFun 1046±17 142 1,534 – – – – Proprietary
27 Qwen2-VL-72BAlibaba開放權重 1047±10 141 5,937 – – – – Qwen
28 Qwen-VL-Max (1119)Alibaba 1058±16 140 1,422 – – – – Proprietary
29 GPT-4.1 nano (2025-04-14)OpenAI 1063±18 139 1,211 – – – – Proprietary
30 GPT-4o (2024-08-06)OpenAI 1065±12 138 3,376 – – – – Proprietary
31 GPT-4o mini (2024-07-18)OpenAI 1066±8 137 17,341 – – – – Proprietary
32 Molmo 2 8BAi2開放權重 1081±20 136 1,190 1058 1072 1069 – Apache 2.0
33 Gemini 1.5 Pro (001)Google 1088±12 135 16,734 – – – – Proprietary
34 Pixtral Large (2411)Mistral AI開放權重 1089±9 134 5,423 – – – – MRL
35 GPT-4 Turbo (2024-04-09)OpenAI 1090±12 133 13,391 – – – – Proprietary
36 Claude 3.5 Haiku (20241022)Anthropic 1093±16 132 1,398 1091 1077 1119 – Proprietary
37 Gemini 2.0 Flash-Lite Preview (02-05)Google 1100±11 131 3,991 – – – – Proprietary
38 Qwen2.5-VL-72B-InstructAlibaba開放權重 1107±11 130 3,768 – – – – Qwen
39 Llama 4 Scout 17B 16E InstructMeta開放權重 1118±9 129 6,446 1117 1125 1148 – Llama
40 Claude 3.5 Sonnet (20240620)Anthropic 1120±10 128 21,624 – – – – Proprietary
41 SStep-1o Vision 32K (High Res)StepFun 1122±13 127 2,833 – – – – Proprietary
42 Claude 3.5 Sonnet (20241022)Anthropic 1126±8 126 10,374 1139 1134 1136 – Proprietary
43 Mistral Small 3.1 24B Instruct (2503)Mistral AI開放權重 1136±8 125 28,490 1132 1119 1158 1153 Apache 2.0
44 GPT-4o (2024-05-13)OpenAI 1137±9 124 23,273 – – – – Proprietary
45 Gemini 1.5 Flash (002)Google 1141±10 123 7,241 – – – – Proprietary
46 Mistral Small 3.2 (2506)Mistral AI開放權重 1142±9 121 10,843 1134 1125 1173 1059 Apache 2.0
47 Llama 4 Maverick 17B 128E InstructMeta開放權重 1142±9 122 6,923 1143 1134 1164 – Llama 4
48 Claude 3.7 Sonnet (20250219)Anthropic 1151±9 120 4,517 1157 1153 1206 – Proprietary
49 Qwen2.5-VL-32B-InstructAlibaba開放權重 1152±15 119 1,490 – – – – Apache 2.0
50 GLM-4.5VZ.ai開放權重 1154±12 118 3,243 1151 1151 1163 – MIT
51 Mistral Medium 3 (2505)Mistral AI 1157±8 117 10,793 1154 1155 1165 1094 Proprietary
52 Gemini 2.0 Flash (001)Google 1158±7 116 10,135 1148 1149 1175 1130 Proprietary
53 GPT-5 nano (High)OpenAI 1159±11 115 3,928 1152 1141 1157 – Proprietary
54 Gemini 1.5 Pro (002)Google 1161±9 114 8,902 – – – – Proprietary
55 GLM-4.6VZ.ai開放權重 1164±14 113 2,455 1171 1182 1155 – MIT
56 Gemma 3 27B ITGoogle開放權重 1165±8 112 16,982 1156 1156 1160 1151 Gemma
57 o1 (2024-12-17)OpenAI 1168±11 111 3,694 – – – – Proprietary
58 Claude 3.7 Sonnet (Thinking 32K, 20250219)Anthropic 1169±16 110 1,436 1181 1176 1192 – Proprietary
59 Mistral Medium 3.1 (2508)Mistral AI 1172±6 109 41,611 1175 1175 1165 1145 Proprietary
60 Claude Sonnet 4 (20250514)Anthropic 1175±14 108 1,806 1173 1178 1215 – Proprietary
61 SStep-3StepFun開放權重 1176±12 106 3,193 1168 1155 1183 – Apache 2.0
62 Claude Opus 4 (20250514)Anthropic 1176±13 107 2,172 1179 1177 1228 – Proprietary
63 Hunyuan Large VisionTencent 1180±16 104 1,274 1167 1135 – – Proprietary
64 Hunyuan Vision 1.5 ThinkingTencent 1180±12 105 2,424 1176 1194 1241 – Proprietary
65 GPT-4.1 mini (2025-04-14)OpenAI 1181±7 103 40,201 1188 1182 1214 1174 Proprietary
66 SStep-1o Turbo (202506)StepFun 1186±14 102 1,895 1177 1185 1155 – Proprietary
67 Gemini 2.5 Flash-Lite Preview (Thinking, 06-17)Google 1187±7 101 35,914 1188 1186 1190 1184 Proprietary
68 Claude Sonnet 4 (Thinking 32K, 20250514)Anthropic 1191±17 100 1,144 1197 1200 – – Proprietary
69 Claude Opus 4 (Thinking 16K, 20250514)Anthropic 1192±16 99 1,290 1197 1179 1197 – Proprietary
70 o4-mini (2025-04-16)OpenAI 1194±7 98 41,316 1197 1197 1234 1204 Proprietary
71 GPT-4.5 Preview (2025-02-27)OpenAI 1195±12 97 2,925 – – – – Proprietary
72 GPT-5.4 nano (High)OpenAI 1196±6 96 25,624 1204 1210 1225 – Proprietary
73 Gemini 2.5 Flash-Lite Preview (Non-Thinking, 09-2025)Google 1198±9 95 4,862 1203 1197 1215 – Proprietary
74 Grok 4.1 Fast (Reasoning)xAI 1201±7 94 13,368 1195 1210 1172 – Proprietary
75 GPT-5 mini (High)OpenAI 1202±8 93 28,943 1199 1202 1210 1217 Proprietary
76 NNVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4)NVIDIA開放權重 1205±13 92 2,509 1216 1223 1229 – OpenMDW-1.1
77 Qwen3-VL-235B-A22B-ThinkingAlibaba開放權重 1207±12 91 2,387 1212 1214 1234 – Apache 2.0
78 GPT-5 (High)OpenAI 1208±7 90 34,445 1215 1227 1227 1204 Proprietary
79 GPT-4.1 (2025-04-14)OpenAI 1210±7 88 40,998 1218 1215 1248 1207 Proprietary
80 Grok 4 (0709)xAI 1210±7 89 30,173 1195 1204 1170 1245 Proprietary
81 Qwen-VL-Max (2025-08-13)Alibaba 1213±12 87 3,170 1236 1268 – – Proprietary
82 o3 (2025-04-16)OpenAI 1214±7 86 45,567 1218 1220 1234 1210 Proprietary
83 GPT-6 Luna (Max)OpenAI 1216±11 85 3,662 1225 1245 1261 – Proprietary
84 Mistral Large 3Mistral AI開放權重 1220±8 84 9,254 1228 1240 1216 – Apache 2.0
85 Mistral Medium 3.5Mistral AI開放權重 1222±9 83 5,375 1233 1254 1230 – Modified MIT
86 MiMo-V2-OmniXiaomi 1228±7 82 10,650 1238 1256 1253 – Proprietary
87 Grok 4.3xAI 1229±6 80 27,901 1235 1241 1241 – Proprietary
88 GPT-5.2OpenAI 1229±6 81 20,100 1233 1246 1258 1174 Proprietary
89 Grok 4.7 (xHigh)xAI 1231±11 79 3,736 1241 1254 1309 – Proprietary
90 GPT-5 ChatOpenAI 1232±7 78 39,894 1246 1252 1278 1212 Proprietary
91 Gemini 2.5 FlashGoogle 1235±5 76 60,087 1237 1245 1244 1277 Proprietary
92 GPT-5.1OpenAI 1235±8 77 10,801 1241 1244 1243 1165 Proprietary
93 TInkling-SmallThinking Machines開放權重 1237±8 75 8,074 1243 1251 1257 – Apache 2.0
94 Gemini 3.1 Flash-Lite PreviewGoogle 1240±6 73 27,657 1246 1242 1271 1259 Proprietary
95 GPT-6 Sol (Max)OpenAI 1240±11 74 3,221 1258 1270 1286 – Proprietary
96 Qwen3.5-27BAlibaba開放權重 1241±5 72 30,272 1246 1249 1256 1170 Apache 2.0
97 ChatGPT-4o Latest (20250326)OpenAI 1243±5 71 23,400 1249 1255 1252 1224 Proprietary
98 Qwen3.5-122B-A10BAlibaba開放權重 1244±7 70 13,236 1251 1249 1261 – Apache 2.0
99 GPT-5.4 mini (High)OpenAI 1245±6 69 25,734 1253 1266 1285 – Proprietary
100 GPT-5.2 (High)OpenAI 1246±6 68 19,022 1257 1265 1276 1267 Proprietary
101 MiMo-V2.5Xiaomi開放權重 1247±6 66 29,381 1256 1269 1262 – MIT
102 Qwen3-VL-235B-A22B-InstructAlibaba開放權重 1247±7 67 12,724 1253 1255 1277 1281 Apache 2.0
103 ERNIE-5.0-Preview (1220)Baidu 1248±11 65 3,349 1250 1236 1253 – Proprietary
104 GPT-5.5 InstantOpenAI 1250±9 63 7,590 1256 1273 1247 – Proprietary
105 GPT-5.1 (High)OpenAI 1250±8 64 9,848 1255 1264 1272 1252 Proprietary
106 MiniMax-M3MiniMax開放權重 1251±6 62 19,844 1259 1266 1268 – MiniMax Community License
107 Kimi K2.5 (Instant)Moonshot AI開放權重 1254±11 60 4,111 1259 1257 1264 – Modified MIT
108 Gemini 2.5 Flash Preview (09-2025)Google 1254±10 61 4,792 1255 1250 1283 – Proprietary
109 MiMo-V2.6-FlashXiaomi開放權重 1256±11 59 3,618 1270 1276 1283 – MIT
110 GPT-5.6 Luna (xHigh)OpenAI 1257±7 58 11,055 1264 1273 1282 – Proprietary
111 Grok 4.20 Multi-Agent Beta (0309)xAI 1259±6 57 24,532 1261 1261 1275 – Proprietary
112 Gemma 4 26B A4BGoogle開放權重 1260±6 56 23,922 1268 1264 1293 1297 Apache 2.0
113 Qwen3.5-397B-A17BAlibaba開放權重 1262±5 54 32,657 1271 1277 1292 1230 Apache 2.0
114 Grok 4.20 Beta (Reasoning, 0309)xAI 1262±6 55 26,796 1266 1267 1269 – Proprietary
115 Gemini 2.5 ProGoogle 1263±5 51 88,755 1270 1274 1281 1304 Proprietary
116 Grok 4.6 (High)xAI 1263±8 52 7,261 1272 1285 1283 – Proprietary
117 MiMo-V2.6-ProXiaomi開放權重 1263±15 53 1,754 1270 1281 1264 – MIT
118 GLM-5V-TurboZ.ai 1264±5 50 42,757 1270 1286 1273 1264 Proprietary
119 SStep 5 PreviewStepFun 1265±15 49 1,649 1282 1308 1265 – Proprietary
120 Gemini 3 Flash (Minimal)Google 1266±5 48 37,173 1272 1276 1298 1241 Proprietary
121 GPT-5.2 Chat (20260210)OpenAI 1268±6 47 16,465 1275 1285 1286 – Proprietary
122 Qwen3.8-27BAlibaba開放權重 1269±8 44 7,614 1281 1285 1295 – Apache 2.0
123 Gemini 3.5 Flash-LiteGoogle 1269±8 45 9,238 1280 1284 1291 – Proprietary
124 Kimi K2.5 (Thinking)Moonshot AI開放權重 1269±6 46 28,249 1276 1282 1297 1257 Modified MIT
125 GPT-5.6 Terra (xHigh)OpenAI 1271±7 43 11,842 1278 1295 1312 – Proprietary
126 Mistral Large 4Mistral AI 1272±19 42 1,008 1284 1288 – – Proprietary
127 Claude Sonnet 5 (High)Anthropic 1273±7 41 14,034 1288 1293 1311 – Proprietary
128 Dola Seed 2.0 ProByteDance 1274±7 40 10,972 1284 1289 1284 – Proprietary
129 Gemma 4 31BGoogle開放權重 1276±6 39 40,219 1282 1283 1310 1218 Apache 2.0
130 Qwen3.7-PlusAlibaba 1278±7 37 14,212 1288 1292 1295 – Proprietary
131 DeepSeek-V4.1-Flash-20260910 (Max)DeepSeek開放權重 1278±10 38 4,100 1282 1297 1307 – MIT
132 GPT-6 Astra (Max)OpenAI 1280±10 36 4,012 1292 1297 1337 – Proprietary
133 GPT-5.6 Sol (xHigh)OpenAI 1281±7 35 11,439 1290 1305 1305 – Proprietary
134 Kimi K2.6Moonshot AI開放權重 1282±7 34 16,368 1291 1298 1310 – Modified MIT
135 Claude Sonnet 4.6Anthropic 1283±6 33 27,408 1296 1306 1311 – Proprietary
136 Gemini 3 FlashGoogle 1285±5 32 39,131 1291 1290 1313 1281 Proprietary
137 Claude Sonnet 5.5 (xHigh)Anthropic 1287±13 30 2,564 1297 1329 1331 – Proprietary
138 Grok 4.5xAI 1287±7 31 12,599 1293 1304 1295 – Proprietary
139 GPT-6.1 Sol (Max)OpenAI 1288±13 29 2,330 1289 1306 1356 – Proprietary
140 Claude Opus 4.8Anthropic 1289±6 28 18,800 1298 1302 1328 – Proprietary
141 GPT-5.4OpenAI 1292±6 27 22,804 1299 1316 1309 – Proprietary
142 GPT-5.5 (High)OpenAI 1293±6 25 27,067 1304 1317 1332 – Proprietary
143 Muse Spark 1.1Meta 1293±7 26 12,804 1301 1305 1287 – Proprietary
144 Claude Opus 4.8 (High)Anthropic 1294±6 23 18,303 1303 1315 1337 – Proprietary
145 GLM-5.3-FlashZ.ai開放權重 1294±6 24 16,254 1299 1317 1318 – MIT
146 Gemini 3.1 Pro PreviewGoogle 1296±5 22 47,114 1304 1311 1316 1345 Proprietary
147 GPT-5.5OpenAI 1297±6 21 28,781 1306 1325 1327 – Proprietary
148 Gemini 3.6 Flash (High)Google 1299±8 20 8,979 1312 1314 1314 – Proprietary
149 Muse Spark 1.3 (xHigh)Meta 1302±10 19 4,897 1309 1318 1296 – Proprietary
150 GPT-5.4 (High)OpenAI 1303±6 18 30,747 1312 1327 1336 – Proprietary
151 Muse Spark 1.2 (xHigh)Meta 1304±14 17 1,948 1313 1317 – – Proprietary
152 Gemini 3 ProGoogle 1305±7 16 13,623 1310 1301 1315 1325 Proprietary
153 Gemini 3.5 Flash (Medium)Google 1306±7 14 14,061 1317 1323 1330 – Proprietary
154 Muse SparkMeta 1306±9 15 5,865 1307 1309 1299 – Proprietary
155 Muse Spark 1.3 (Max)Meta 1308±9 12 5,392 1319 1330 1296 – Proprietary
156 Gemini 3.5 Flash (High)Google 1308±7 13 13,862 1312 1318 1321 – Proprietary
157 Claude Opus 4.6Anthropic 1312±6 11 26,841 1326 1340 1327 – Proprietary
158 Claude Opus 4.7 (High)Anthropic 1313±7 9 22,731 1320 1336 1337 – Proprietary
159 Qwen3.8-MaxAlibaba 1313±7 10 9,957 1322 1323 1336 – Proprietary
160 Gemini 3.7 Flash (High)Google 1314±9 8 5,369 1322 1329 1320 – Proprietary
161 Gemini 3.8 Flash (High)Google 1315±10 7 4,645 1333 1345 1351 – Proprietary
162 Claude Opus 4.6 (High)Anthropic 1316±6 6 22,286 1326 1332 1337 – Proprietary
163 Claude Opus 4.7Anthropic 1317±7 5 23,139 1327 1336 1335 – Proprietary
164 Claude Fable 5.1 (Max)Anthropic 1319±9 4 5,026 1327 1343 1345 – Proprietary
165 Claude Opus 5 (High)Anthropic 1320±7 3 17,743 1329 1348 1338 – Proprietary
166 Claude Opus 5.5 (High)Anthropic 1321±12 2 3,035 1326 1345 1362 – Proprietary
167 Claude Fable 5 (High)Anthropic 1325±7 1 14,714 1337 1357 1361 – Proprietary

得分資料：LMArena leaderboard-dataset（lmarena-ai，Hugging Face，CC-BY-4.0，vision 設定），綜合發布於 2026年10月19日。引用請註明來源。

資料來源與計算方法

  1. 得分為 Bradley-Terry 模型擬合的 Arena 分數，來自真人在同一圖片問題下對兩個匿名回答的投票。
  2. 分項（OCR、圖表、作業題、圖片描述、中文／英文等）是 LMArena 依提示詞類別單獨計算的分榜，不是綜合分的拆分，名次可能和綜合榜差很多。
  3. 沒有公開資料的欄位（價格、上下文、發布時間等）不顯示，空白為 –；不做推測。
  4. 每天自動同步一次；JSON 介面： /api/public/vision-leaderboard.php

本週變化

資料累積中：已記錄 1 天，滿 7 天後顯示每週變化

每天自動同步一次並記錄名次，只用真實快照對比，不回填。

常見問題

目前最強的多模態（視覺）大模型是哪個？

依 LMArena 視覺競技場 2026年10月19日 發布的資料，Anthropic 的 Claude Fable 5 (High) 以 1325 分（14,714 票）在 167 個模型中排第一。

各家廠商最好的視覺模型分別是哪個？

綜合榜各廠商最高：Anthropic Claude Fable 5 (High) (1325, #1)、Google Gemini 3.8 Flash (High) (1315, #7)、Alibaba Qwen3.8-Max (1313, #10)、Meta Muse Spark 1.3 (Max) (1308, #12)、OpenAI GPT-5.4 (High) (1303, #18)。

哪個模型 OCR 文字辨識最好？

LMArena OCR 分榜（123 個模型）第一是 Claude Fable 5 (High)，1337 分。

最好的開源／開放權重視覺模型是哪個？

榜上有 50 個開放權重模型，最高的是 GLM-5.3-Flash（1294 分，授權 MIT）。

這個分數和 MMMU 等基準測試有什麼區別？

MMMU 等是固定題庫的自動評分；這裡是真人用自己的圖片和問題做盲測投票，反映實際使用中的偏好，兩者不能直接換算。