General
OSWorld
OSWorld tests whether agents can complete real tasks on an Ubuntu desktop driving applications such as Chrome, LibreOffice and VS Code, with each task graded by a scorer that can award partial credit.
369items
67subjects
Apache-2.0license
gui_agentdomain
textmodality
gui_screenshotmodality
item-level responses released
Saturation status: No
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
lowhighUnobserved
Scale: [0, 1] per-task rubric scorer output