# Arena vision style control / captioning > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acaptioning. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acaptioning. Human pairwise preferences on captioning prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response style; these are preference ratings rather than task accuracy. Canonical URL: https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acaptioning Last updated: Wed, 21 Oct 2026 18:52:20 GMT Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. Human pairwise preferences on captioning prompts in Arena’s vision style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response style; these are preference ratings rather than task accuracy. Multimodal · Models tested: 22 · Score publishers: [Arena](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset) ## Model results | # | Model | Score | | | --- | --- | --- | --- | | 1 | [Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview.md) | 1,281.62 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1281.6156265182872 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1233.305276376888 rank 1 upper 1329.925976659686 variance 607.5530260045458 votes 137 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 3.1 Pro Preview Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/167/row` | | | | | 2 | [Gemini 3 Pro](https://benchize.com/models/gemini-3-pro.md) | 1,276.59 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1276.5850642680869 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1239.4042985800359 rank 2 upper 1313.7658299561376 variance 359.86571812318823 votes 218 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 3 Pro Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/168/row` | | | | | 3 | [GPT-5.2](https://benchize.com/models/gpt-5-2-high.md)High | 1,262.12 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1262.1193656790242 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control Reasoning effort high thinkingEnabled true ### Reported statistics lower 1204.614467044801 rank 3 upper 1319.6242643132475 variance 860.8222868662131 votes 94 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5.2 (High) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/169/row` | | | | | 4 | [Gemini 2.5 Pro](https://benchize.com/models/gemini-2-5-pro.md) | 1,250.59 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1250.5863934082188 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1227.125807419521 rank 4 upper 1274.0469793969164 variance 143.27866589850018 votes 796 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 2.5 Pro Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/170/row` | | | | | 5 | [GPT-5.1](https://benchize.com/models/gpt-5-1-high.md)High | 1,248.24 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1248.2369734748138 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control Reasoning effort high thinkingEnabled true ### Reported statistics lower 1193.0120981990924 rank 5 upper 1303.461848750535 variance 793.9137165260296 votes 101 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5.1 (High) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/171/row` | | | | | 6 | [Gemma 4 26B A4B](https://benchize.com/models/gemma-4-26b-a4b-it.md) | 1,238.93 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1238.9268344996578 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1167.6024291227477 rank 6 upper 1310.251239876568 variance 1324.2809671588718 votes 63 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemma 4 26B A4B Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/172/row` | | | | | 7 | [Gemini 3 Flash](https://benchize.com/models/gemini-3-flash.md) | 1,233.79 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1233.7928057263364 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1194.6685546171072 rank 7 upper 1272.9170568355655 variance 398.4702417248417 votes 191 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 3 Flash Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/173/row` | | | | | 8 | [Gemini 3.1 Flash Lite Preview](https://benchize.com/models/gemini-3-1-flash-lite-preview.md) | 1,225.48 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1225.4806546493785 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1161.5229704972508 rank 8 upper 1289.438338801506 variance 1064.852066112784 votes 73 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 3.1 Flash-Lite Preview Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/174/row` | | | | | 9 | [Gemini 2.5 Flash](https://benchize.com/models/gemini-2-5-flash.md) | 1,211.35 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1211.3500947836817 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1185.5942275852256 rank 10 upper 1237.105961982138 variance 172.68561921604896 votes 582 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 2.5 Flash Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/176/row` | | | | | 10 | [Qwen3 VL 235B A22B Instruct](https://benchize.com/models/qwen3-vl-235b-a22b-instruct.md) | 1,211.03 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1211.0336304822602 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1165.502795876151 rank 11 upper 1256.5644650883694 variance 539.6535526454735 votes 146 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Qwen3-VL-235B-A22B-Instruct Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/177/row` | | | | | 11 | [Gemini 3 Flash](https://benchize.com/models/gemini-3-flash-minimal.md)Minimal | 1,209.01 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1209.0135138536432 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control Reasoning effort minimal thinkingEnabled true ### Reported statistics lower 1165.8963556919568 rank 13 upper 1252.1306720153293 variance 483.95399110485215 votes 173 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 3 Flash (Minimal) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/179/row` | | | | | 12 | [GPT-5](https://benchize.com/models/gpt-5-high.md)High | 1,196.05 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1196.0498457069082 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control Reasoning effort high thinkingEnabled true ### Reported statistics lower 1159.5589268077715 rank 15 upper 1232.540764606045 variance 346.63580276587805 votes 356 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5 (High) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/181/row` | | | | | 13 | [GPT-5 Chat](https://benchize.com/models/gpt-5-chat.md) | 1,194.22 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1194.223307795897 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1162.111763052961 rank 16 upper 1226.3348525388333 variance 268.4270101302992 votes 375 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5 Chat Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/182/row` | | | | | 14 | [GPT-5 Mini](https://benchize.com/models/gpt-5-mini-high.md)High | 1,175.17 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1175.1710880182416 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control Reasoning effort high thinkingEnabled true ### Reported statistics lower 1132.3016498833047 rank 22 upper 1218.0405261531785 variance 478.4090658749028 votes 283 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5 mini (High) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/188/row` | | | | | 15 | [Qwen3.5 397B A17B](https://benchize.com/models/qwen3-5-397b-a17b.md) | 1,174.48 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1174.4848521263846 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1117.2431604366516 rank 23 upper 1231.7265438161176 variance 852.9601436441734 votes 78 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Qwen3.5-397B-A17B Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/189/row` | | | | | 16 | [GPT-5.2](https://benchize.com/models/gpt-5-2.md) | 1,171.86 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1171.8626560483087 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1116.2310214530714 rank 24 upper 1227.494290643546 variance 805.6519442732898 votes 107 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5.2 Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/190/row` | | | | | 17 | [Gemma 4 31B](https://benchize.com/models/gemma-4-31b-it.md) | 1,170.08 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1170.0801426299788 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1115.073951529153 rank 26 upper 1225.0863337308047 variance 787.6385510424009 votes 106 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemma 4 31B Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/192/row` | | | | | 18 | [GPT-5.1](https://benchize.com/models/gpt-5-1.md) | 1,162.65 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1162.6466525959077 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1114.3328437825835 rank 27 upper 1210.9604614092318 variance 607.6400219301732 votes 123 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label GPT-5.1 Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/193/row` | | | | | 19 | [Gemini 2.0 Flash 001](https://benchize.com/models/gemini-2-0-flash-001.md) | 1,149.87 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1149.866231814507 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1089.3922042944623 rank 28 upper 1210.3402593345515 variance 952.0102063294524 votes 106 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemini 2.0 Flash (001) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/194/row` | | | | | 20 | [Mistral Small 3.1 24B](https://benchize.com/models/mistral-small-3-1-24b-instruct.md) | 1,133.08 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1133.0815402746682 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1094.1270215382353 rank 29 upper 1172.0360590111009 variance 395.02038179154624 votes 258 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Mistral Small 3.1 24B Instruct (2503) Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/195/row` | | | | | 21 | [Gemma 3 27B](https://benchize.com/models/gemma-3-27b-it.md) | 1,116.75 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1116.7484194373712 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1077.7485995319785 rank 31 upper 1155.7482393427642 variance 395.93967387062725 votes 258 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Gemma 3 27B IT Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/197/row` | | | | | 22 | [Qwen3.5-27B](https://benchize.com/models/qwen3-5-27b.md) | 1,095.45 | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `rating` \| 1095.4516885749845 \| rating \| reported \| Higher is better \| ### Configuration category captioning config vision_style_control ### Reported statistics lower 1012.9713618936328 rank 32 upper 1177.932015256336 variance 1770.942917001833 votes 61 ### Source [Original source](https://datasets-server.huggingface.co/parquet/lmarena-ai%2Fleaderboard-dataset/vision_style_control/latest) Source Arena Original model label Qwen3.5-27B Observed 2026-10-21T18:45:05.190Z Evidence SHA-256 `38568936995f36625aae49e9ea05557d55c1b1df9b2954f5c50cee523f26fe92` Source locator `/rows/198/row` | | | | --- Cite as: Arena vision style control / captioning — Benchize, https://benchize.com/benchmarks/lmarena%3Avision_style_control%3Acaptioning (retrieved 2026-10-21)