Atlas

Benchmarks

← All benchmarks

Arena Code WebDev — Reference-Based Design

Code

Arena Code WebDev rating for reference-based design tasks. Anonymous generated experiences are compared through blind pairwise user votes.

Top models (higher is better)

ModelScore
Claude Opus 5.51837
Claude Sonnet 5.51824
GPT-6 Astra1807
Claude Fable 5.11800
GPT-6.1 Sol1779
Muse Wonder Full1725
GPT-6 Sol1720
Claude Opus 51708
Gemini 4 Argon1703
Qwen3.8 Max (0902)1688
Qwen3.8-Max1685
Kimi K3.1 Pro1684
Kimi K31679
Muse Spark 1.31672
DeepSeek V4.1 Flash 10171669
Muse Wonder1660
Claude Fable 51655
GPT-5.6 Sol1653
Grok 4.71651
Hy4 Preview1650
MiMo-V2.6-Pro1644
MiMo-V2.6-Flash1644
GLM-5.31640
Grok 4.61635
MiniMax M3.1 Flash1635
Qwen3.8-Flash-Next1631
GLM-5.3 Flash1615
DeepSeek V4.1 Flash1614
Gemini 3.7 Flash1608
GLM-5.21608
Claude Haiku 5.51606
Qwen3.8 27B1597
Gemini 3.8 Flash1596
GPT-6 Luna1587
DeepSeek V4 Pro 08131583
Opus 4.71580
DeepSeek-V4-Flash1577
Opus 4.81577
Grok 4.51575
Opus 4.61561
Loading Atlas data…