LMArena Vision: captioning leaderboard
- models measured
- 23
- lead over #2
- +41 Elo
- median
- 1218
Leader on LMArena Vision: captioning · read October 22
On October 22, 2026, 24 of our catalog's models had a LMArena Vision: captioning result, and Google Gemini 3.1 Pro Preview led with 1345, ahead of Google Gemini 2.5 Pro at 1304.
Image prompts asking for a caption. People ask a question, compare two anonymous answers and vote for the better one; the rating is built from those votes, as LMArena publishes it.
Role in the index
LMArena Vision: captioning is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Votes | Interval (±) | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro PreviewGoogle | 1345 | 137 | ±48.5 | 32.3 | $12.00 | |
| 2 | Gemini 2.5 ProGoogle | 1304 | 796 | ±22.6 | 19.1 | $10.00 | |
| 3 | Gemma 4 26B A4BGoogle | 1297 | 63 | ±69.8 | Not benchmarked | $0.60 | |
| 4 | Qwen3 VL 235B A22B InstructQwen | 1281 | 146 | ±44.5 | Not benchmarked | $1.90 | |
| 5 | Gemini 2.5 FlashGoogle | 1277 | 582 | ±24.4 | Not benchmarked | $2.50 | |
| 6 | GPT-5.2OpenAI | 1267 | 94 | ±58.0 | 55.9 | $14.00 | |
| 7 | GLM 5V TurboZ.ai | 1264 | 53 | ±72.7 | Not benchmarked | $4.00 | |
| 8 | Gemini 3.1 Flash Lite PreviewGoogle | 1259 | 73 | ±64.5 | 22.7 | $1.50 | |
| 9 | Kimi K2.5Moonshot AI | 1257 | 125 | ±48.2 | 38.6 | $2.85 | |
| 10 | GPT-5.1OpenAI | 1252 | 101 | ±55.3 | 40.4 | $10.00 | |
| 11 | Qwen3.5 397B A17BQwen | 1230 | 78 | ±56.7 | Not benchmarked | $2.34 | |
| 12 | Gemma 4 31BGoogle | 1218 | 106 | ±54.1 | 39.0 | $0.40 | |
| 13 | GPT-5 MiniOpenAI | 1217 | 283 | ±36.5 | 34.2 | $2.00 | |
| 14 | GPT-5OpenAI | 1212 | 375 | ±31.7 | 38.8 | $10.00 | |
| 15 | o3OpenAI | 1210 | 540 | ±27.6 | Not benchmarked | $8.00 | |
| 16 | GPT-4.1OpenAI | 1207 | 422 | ±31.1 | Not benchmarked | $8.00 | |
| 17 | o4 MiniOpenAI | 1204 | 423 | ±29.7 | Not benchmarked | $4.40 | |
| 18 | GPT-4.1 MiniOpenAI | 1174 | 394 | ±32.3 | Not benchmarked | $1.60 | |
| 19 | Qwen3.5-27BQwen | 1170 | 61 | ±81.0 | Not benchmarked | $1.56 | |
| 20 | Mistral Small 3.1 24BMistral AI | 1153 | 258 | ±38.5 | Not benchmarked | $0.555 | |
| 21 | Gemma 3 27BGoogle | 1151 | 258 | ±38.6 | Not benchmarked | $0.30 | |
| 22 | Mistral Medium 3.1Mistral AI | 1145 | 403 | ±30.3 | Not benchmarked | $2.00 | |
| 23 | Gemini 2.0 FlashRetiredGoogle | 1130 | 106 | ±60.9 | Not benchmarked | $0.40 | |
| 24 | Mistral Medium 3Mistral AI | 1094 | 160 | ±49.2 | Not benchmarked | $2.00 |
Only models in our catalog are listed, and each appears once with its best published run. 184 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Every other column is a figure the board publishes for the same run.