Arena WebDev
LMArena · Anonymous human comparisons assess whether generated websites are usable and ready to deliver.
What it measures and how
Only WebDev Overall in official snapshots is used. The live arena.ai page is not scraped. It shares one Arena family vote with Text.
How this evidence is used
Receives 2.4% of the coding-and-design budget. It measures website design preferences, not code correctness, and shares an evidence family with other Arena tasks.
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 | Claude Opus 5.5 (Max)Anthropic | 1,807.3 | Max reasoning |
| 2 | GPT-6 Astra (Max)OpenAI | 1,777.2 | Max reasoning |
| 3 | Claude Sonnet 5.5 (xHigh)anthropic | 1,767.6 | xHigh reasoning |
| 4 | GPT-6.1 Sol (Max)openai | 1,755.5 | Max reasoning |
| 5 | Claude Fable 5.1 (Max)Anthropic | 1,741.4 | Max reasoning |
| 7 | Claude Opus 5 (Max)Anthropic | 1,689.3 | Max reasoning |
| 8 | GPT-6 Sol (Max)OpenAI | 1,682.7 | Max reasoning |
| 9 | Gemini 4 Argon (High)google | 1,678.2 | High reasoning |
| 11 | Qwen3.8-MaxAlibaba | 1,672.3 | Source default |
| 12 | Muse Spark 1.3 (Max)Meta | 1,658.4 | Max reasoning |
| 14 | Kimi K3 (Max)Moonshot AI | 1,655.2 | Max reasoning |
| 15 | Muse Wonder (Max)meta | 1,650.4 | Max reasoning |
| 16 | Grok 4.7 (xHigh)xAI | 1,638.7 | xHigh reasoning |
| 17 | MiMo-V2.6-FlashXiaomi | 1,634.3 | Source default |
| 18 | Hy4 previewTencent | 1,631.5 | Source default |
| 19 | Qwen3.8-Flash-NextAlibaba | 1,630.3 | Source default |
| 20 | MiMo-V2.6-ProXiaomi | 1,624.8 | Source default |
| 21 | Claude Fable 5 (High)Anthropic | 1,624.7 | High reasoning |
| 23 | GLM-5.3 (Max)Z.ai | 1,621.1 | Max reasoning |
| 24 | DeepSeek-V4.1-Flash (Max)DeepSeek | 1,618.4 | Max reasoning |
| 25 | Grok 4.6 (High)xAI | 1,617.4 | High reasoning |
| 26 | GPT-5.6 Sol (xHigh, Codex Harness)OpenAI | 1,616.3 | xHigh reasoning · Codex Harness |
| 27 | GLM-5.3-FlashZ.ai | 1,605.5 | Source default |
| 28 | GLM-5.2 (Max)Z.ai | 1,604.6 | Max reasoning |
| 29 | Qwen3.8-27BAlibaba | 1,593.2 | Source default |
| 30 | Gemini 3.7 Flash (High)Google | 1,590.6 | High reasoning |
| 31 | Gemini 3.8 Flash (High)Google | 1,585.2 | High reasoning |
| 32 | Claude Haiku 5.5 (High)— | 1,584.7 | High reasoning |
| 33 | DeepSeek-V4-Pro (High, 20260813)DeepSeek | 1,582 | High reasoning |
| 34 | DeepSeek-V4-Flash (High)DeepSeek | 1,581.4 | High reasoning |
Evaluation limits & attribution
Its preference construct is correlated with Text.
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.