LMArena WebDev
2026-10-08 · Frontend
- What it measures
- Frontend web work: landing pages, interactive components, small games, and full pages, including agent flows that need several steps and tool calls.
- How
- A person submits a brief. Two anonymous models each build an interactive web app and run it in a sandbox. The person votes after trying them. Only votes cast before the identities are revealed count.
- What the score means
- A Bradley-Terry model turns pairwise votes into a strength score, read a bit like chess Elo: 100 points is about a 64% win rate. ± is the 95% interval. The band under the rank is the statistically plausible rank range.
- Caveats
- It measures human preference (does it look right, does it run), not test pass rate. Briefs are mostly single-page demos, so it says little about large codebases or long refactors. Effort tiers of the same model are ranked separately. Preliminary rows have few votes and can move.
- Updates
- Updated continuously. This page's board date is 2026-10-08, synced 2026-10-10. We sync weekly and write changes on the home page.
Full board
142
All
| # | Rank spread | Model | Score | CI | Votes | Price $/M | Context |
|---|