Skip to content

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

Agents and tools · University of Hong Kong (XLang) · introduced Apr 11, 2024 · Success rate

About this benchmark

Task
Operate a real computer (Ubuntu, Windows, macOS) through screenshots and input to finish open-ended tasks.
Dataset
369 tasks with real web and desktop apps, file I/O and multi-app workflows.
Method
Execution-based checking scripts per task. People completed 72%, the best model 12% at release.
Metric
Success rate
Organization
University of Hong Kong (XLang)
Introduced
Apr 11, 2024
Official leaderboard
os-world.github.io/

Versions

Newest first. New versions are added, never rewritten.

VersionDate
1Released with the paper.Apr 11, 20242 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.