OSWorld 2.0 evaluates computer-use agents on 108 long-horizon, stateful workflows spanning everyday and professional applications.
Captures multi-hour, cross-application workflows with dynamic state and implicit information. Tests visual-spatial precision and constraint tracking that short OSWorld 1.x tasks underrepresent.
The task set, horizon, step budget, binary-completion metric, and partial score are not comparable with OSWorld-Verified. Preserve each metric and system setup separately.
Higher is better. Report binary task completion separately from partial score.
Max score: 100
The task set, horizon, step budget, binary-completion metric, and partial score are not comparable with OSWorld-Verified.
No scores available for this benchmark yet.