Benchmarks › WebDev Arena

WebDev Arena

By LMArena / Arena ↗Since 2024
Build an interactive web app from a user prompt
1,807top score · Claude Opus 5.5
97models ranked
46open-weights models
Bradley-Terry / Elo-style preference ratinghigher is better

How the test works

1

Prompts

Organic user prompts

  • What the model doesBuild an interactive web app from a user prompt
  • Kinds of tasksWeb application generation
  • In → outText → Web application
2

Votes

923,854 votes

  • Graded byHuman pairwise votes
  • How runs combineBradley-Terry fit to human pairwise votes, reported on an Elo-like scale
3

Rating

Bradley-Terry / Elo-style preference rating

  • Better isHigher
Sources: arena.ai

WebDev Arena leaderboardtop 10 of 97

Full leaderboard with filters →
ModelWebDev ArenaPer taskPriceSpeedSupportsContext
Claude Opus 5.5Anthropic
1,807max
$5.98$4 / $20961M
GPT-6 AstraOpenAI
1,774max
$3.26$10 / $50691.05M
Claude Sonnet 5.5Anthropic
1,768xhigh
$2.01$2 / $101121M
GPT-6.1 SolOpenAI
1,753max
$0.72$2 / $10521.05M
Claude Fable 5.1Anthropic
1,743max
$7.63$10 / $50691M
Muse Wonder FullMeta
1,726
——1M
Claude Opus 5Anthropic
1,689max
$5.86$5 / $25551M
GPT-6 SolOpenAI
1,683max
$1.04$2 / $101281.05M
Gemini 4 ArgonGoogle
≈1,678high
$1.99$2 / $10841M
Qwen3.8 Max (0902)Alibaba (Qwen)
≈1,674max
—$2 / $6—984K

WebDev Arena questions

What is WebDev Arena?›

Build an interactive web app from a user prompt. Created by LMArena / Arena.

Which model is best on WebDev Arena?›

Claude Opus 5.5 leads WebDev Arena with 1,807, ahead of GPT-6 Astra (1,774) and Claude Sonnet 5.5 (1,768).

What is the best open-weights model on WebDev Arena?›

DeepSeek V4.1 Flash 1017 with 1,664, then Kimi K3 (1,655).

How is WebDev Arena scored?›

Bradley-Terry / Elo-style preference rating; human pairwise votes.