WebDev Arena (Reference-Based Design): leaderboard
Voters submit a prompt for a web app, two anonymous models build it, and both builds run in sandboxes the voter can click through before choosing one. This board counts only prompts tagged reference-based design, which ask for a site modelled on a known product or interface and its recognizable layout and style. Votes are fitted with a Bradley-Terry model and higher is better.
Metric: Arena Score. Source: arena.ai. 141 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Max) | 1853 | #3 (Claude Opus 5.5) |
| 2 | Claude Sonnet 5.5 (xHigh) | 1826 | #11 (Claude Sonnet 5.5) |
| 3 | GPT-6 Astra (Max) | 1821 | #6 (GPT-6 Astra) |
| 4 | Claude Fable 5.1 (Max) | 1801 | #9 (Claude Fable 5.1) |
| 5 | GPT-6.1 Sol (Max) | 1788 | #7 (GPT-6.1 Sol) |
| 6 | Claude Sonnet 5.5 (High) | 1776 | #11 (Claude Sonnet 5.5) |
| 7 | GPT-6 Sol (Max) | 1734 | #14 (GPT-6 Sol) |
| 8 | Claude Opus 5 (Max) | 1710 | #12 (Claude Opus 5) |
| 9 | Gemini 4 Argon (High) | 1704 | #8 (Gemini 4 Argon) |
| 10 | Qwen 3.8 Max | 1684 | #35 |
| 11 | Claude Opus 5 (High) | 1683 | #12 (Claude Opus 5) |
| 12 | Qwen 3.8 Max (0902) | 1683 | #29 |
| 13 | Kimi K3 (Max) | 1677 | #30 (Kimi K3) |
| 14 | Muse Spark 1.3 (Max) | 1666 | #24 (Muse Spark 1.3) |
| 15 | GPT-5.6 Sol (xHigh) | 1655 | #19 (GPT-5.6 Sol) |
Interactive version: theaggregate.ai/benchmark?slug=webdev-arena-reference-based-design · How It Works · Data refreshed daily, snapshot 2026-10-20.