Arena Text Style-Controlled
LMArena · People anonymously compare answer pairs, adding an overall experience measure that multiple-choice tests miss.
What it measures and how
Only Text Overall style-controlled in official published snapshots is used. The live arena.ai page is not scraped.
How this evidence is used
The 10% human-preference budget splits into 5% overall and 5% creative writing. It shares an evidence family with Arena coding and vision. Freshness uses the official results date.
Evaluation results
Fixed rules select one representative configuration per public model. Anonymous test models are omitted. Original ranks are retained, with up to 30 models per evaluation.
| Source rank | Source model name | Raw result | Representative configuration |
|---|---|---|---|
| 1 | Gemini 4 Argon (High)google | 1,523.9 | High reasoning |
| 2 | Muse Wonder (Max)meta | 1,508.9 | Max reasoning |
| 3 | Claude Opus 5.5 (High)Anthropic | 1,508.2 | High reasoning |
| 4 | Claude Fable 5 (High)Anthropic | 1,503.6 | High reasoning |
| 5 | Claude Opus 4.6 (High)Anthropic | 1,503.1 | High reasoning |
| 6 | Claude Fable 5.1 (Max)Anthropic | 1,502.3 | Max reasoning |
| 7 | Claude Opus 4.7 (High)Anthropic | 1,501.1 | High reasoning |
| 9 | Gemini 3.8 Flash (High)Google | 1,496.4 | High reasoning |
| 10 | Muse Spark 1.3 (Max)Meta | 1,494.4 | Max reasoning |
| 12 | Muse Spark 1.2 (xHigh)Meta | 1,493.1 | xHigh reasoning |
| 13 | Muse Spark 1.1Meta | 1,490.2 | Source default |
| 14 | Claude Opus 5 (Max)Anthropic | 1,490 | Max reasoning |
| 16 | Muse SparkMeta | 1,489 | Source default |
| 17 | Gemini 3.7 Flash (High)Google | 1,487.6 | High reasoning |
| 18 | Kimi K3 (Max)Moonshot AI | 1,487.1 | Max reasoning |
| 19 | Gemini 3.1 Pro PreviewGoogle | 1,486.9 | Source default |
| 20 | Gemini 3 ProGoogle | 1,485.5 | Source default |
| 21 | gpt-5.6-sol-xhighOpenAI | 1,484.7 | xHigh reasoning |
| 22 | gpt-6.1-sol-maxopenai | 1,483.8 | Max reasoning |
| 23 | gemini-3.6-flash-highGoogle | 1,482.7 | High reasoning |
| 24 | qwen3.8-maxAlibaba | 1,482.4 | Source default |
| 25 | claude-opus-4-8-highAnthropic | 1,481.5 | High reasoning |
| 26 | gpt-5.5-highOpenAI | 1,481.4 | High reasoning |
| 27 | mimo-v2.6-proXiaomi | 1,479.2 | Source default |
| 28 | glm-5.3-maxZ.ai | 1,478.5 | Source default |
| 29 | gemini-3.5-flash-highGoogle | 1,477.5 | High reasoning |
| 30 | gpt-6-astra-maxOpenAI | 1,477.3 | Max reasoning |
| 32 | glm-5.2-maxZ.ai | 1,475.9 | Source default |
| 34 | gpt-5.2-chat-latest-20260210OpenAI | 1,475.6 | Source default |
| 35 | gpt-5.4-highOpenAI | 1,475.1 | High reasoning |
Evaluation limits & attribution
Human preferences depend on style, users, and prompts. Overall and creative-writing categories have overlapping votes and share the original human-preference budget, rather than representing independent evaluators.
Data license: CC BY 4.0
Results are published by LMArena. Original scores and the NexGateHub consensus index use different scales and cannot be added directly.