Skip to content

WebArena what it measures, how it is scored and who leads it

Agents and tools · Carnegie Mellon University · introduced Jul 25, 2023 · Task success

About this benchmark

Task
Complete tasks on realistic self-hosted websites: shopping, forums, code hosting, maps, content management.
Dataset
Long-horizon tasks on functional copies of popular site types.
Method
Programmatic checks of the final site state or answer. People reached 78%, the best GPT-4 agent 14% at release.
Metric
Task success
Organization
Carnegie Mellon University
Introduced
Jul 25, 2023

Versions

Newest first. New versions are added, never rewritten.

VersionDate
1Released with the paper.Jul 25, 20233 years ago

Sources: each benchmark's paper (arXiv) and its official site or leaderboard. Scores are the boards captured here every day, exactly as published.

New #1s by email

Saturdays, only in weeks when a leaderboard has a new #1.

Double opt-in. Unsubscribe any time.