Arena Vision Style-Controlled
LMArena · Humans blindly select answers with images to observe the actual experience of visual understanding.
What it measures, and how
Only the overall of the official dataset vision_style_control is read; image generation and image editing are not included.
How this evidence is used
Visual budget 6%; this is still human-preference evidence and cannot be equated with strict image accuracy.
Results
Under fixed rules, each public model uses one representative configuration. Anonymous test variants are not shown, original leaderboard ranks are kept, and each item lists at most 30.
| Rank there | Model there | Raw score | Representative configuration |
|---|---|---|---|
| 1 | Claude Fable 5 (High)Anthropic | 1,309.4 | High reasoning |
| 3 | Qwen3.8-MaxAlibaba | 1,299 | Source default configuration |
| 4 | Claude Opus 4.6 (High)Anthropic | 1,298.8 | High reasoning |
| 5 | Claude Opus 4.7 (High)Anthropic | 1,298.4 | High reasoning |
| 6 | Claude Opus 5.5 (High)Anthropic | 1,295.3 | High reasoning |
| 7 | Muse SparkMeta | 1,294.4 | Source default configuration |
| 8 | Gemini 3.7 Flash (High)Google | 1,293.8 | High reasoning |
| 9 | Muse Spark 1.2 (xHigh)Meta | 1,293.5 | xHigh reasoning |
| 11 | Gemini 3.8 Flash (High)Google | 1,293.1 | High reasoning |
| 12 | GPT-6.1 Sol (Max)— | 1,292.5 | Max reasoning |
| 13 | Muse Spark 1.3 (Max)Meta | 1,291.5 | Max reasoning |
| 14 | Gemini 3 ProGoogle | 1,289.4 | Source default configuration |
| 15 | Claude Fable 5.1 (Max)Anthropic | 1,289.3 | Max reasoning |
| 16 | GPT-5.6 Sol (xHigh)OpenAI | 1,288 | xHigh reasoning |
| 18 | Claude Opus 4.8 (High)Anthropic | 1,285.9 | High reasoning |
| 19 | Claude Opus 5 (High)Anthropic | 1,285.4 | High reasoning |
| 21 | GPT-6 Astra (Max)OpenAI | 1,283.4 | Max reasoning |
| 22 | GPT-5.5 (High)OpenAI | 1,283.3 | High reasoning |
| 23 | GPT-5.4 (High)OpenAI | 1,282.2 | High reasoning |
| 25 | Gemini 3.5 Flash (High)Google | 1,281.1 | High reasoning |
| 26 | Grok 4.5xAI | 1,280.4 | Source default configuration |
| 27 | Gemini 3.1 Pro PreviewGoogle | 1,279.9 | Source default configuration |
| 28 | Gemini 3.6 Flash (High)Google | 1,279.8 | High reasoning |
| 29 | Muse Spark 1.1Meta | 1,279.3 | Source default configuration |
| 32 | GPT-5.2 Chat (20260210)openai | 1,277.7 | Source default configuration |
| 33 | GPT-5.5 InstantOpenAI | 1,277.4 | Source default configuration |
| 34 | Claude Sonnet 4.6Anthropic | 1,274.9 | Source default configuration |
| 35 | Gemini 3 FlashGoogle | 1,272.2 | Source default configuration |
| 36 | GLM-5.3-FlashZ.ai | 1,271.9 | Source default configuration |
| 37 | GPT-5.6 Terra (xHigh)OpenAI | 1,266.4 | xHigh reasoning |
Limits and data attribution
Human preference is shaped by style, user composition and prompt distribution. Overall and creative categories overlap in voting and share the existing human-preference budget, so they are not treated as two independent evaluators.
Data licence: CC BY 4.0
Scores published by LMArena. Raw scores and the News consensus score use different scales and cannot be added directly.