Atlas

Benchmarks

← All benchmarks

Arena Vision — Captioning (No Style Control)

Multimodal

Raw Vision Arena rating for image-captioning prompts without style-control adjustment. Higher ratings indicate stronger preference in blind pairwise user votes.

Top models (higher is better)

ModelScore
Gemini 3.1 Pro Preview1336
Gemini 3 Pro Preview1316
Gemini 2.5 Pro1303
Gemma 4 26B A4B1301
Gemini 2.5 Flash1278
GPT-5.21274
Gemini 3 Flash Preview1272
Gemini 3.1 Flash-Lite Preview1258
Kimi K2.51252
Grok 41244
GPT-5.11243
GPT-5 Mini1211
O3 (2025-04-16)1210
Gemma 4 31B1208
GPT-4.1 (2025-04-14)1204
O4 Mini (2025-04-16)1202
GPT-51201
Gemini 2.5 Flash-Lite Preview 06-171186
GPT-4.1 Mini (2025-04-14)1176
Mistral Small 3.1 24B Instruct 25031154
Gemma 3 27B IT1151
Mistral Medium 3.11144
Gemini 2.0 Flash 0011127
Mistral Medium 31097
Mistral Small 3.2 24B Instruct 25061063
Loading Atlas data…