0

computer use

Slug
computer-use
Evals
2
Models
9
Papers
2

Evals testing this capability

2
View all

OSWorld-Verified

XLANG Lab

Cleaned, human-validated subset of OSWorld tasks designed for stable cross-lab comparison of computer-use agents.

ActiveComputer UsePlanningTool CallingAgentic

OSWorld

XLANG Lab

369 computer-use tasks across Ubuntu, Windows, and macOS environments testing whether agents can operate a real desktop via screenshots and mouse/keyboard.

ActiveComputer UsePlanningTool CallingAgentic

Tools lifting evals here

View all
No tools yet target evals in this capability.

Top models on this capability

9

by avg parsed score across evals here

computer useBar chart with 9 bars. Highest value: Kimi K2.6 at 73.1.
9 models

Papers in this area

2
introducesOSWorld-Verified: A Cleaner, More Reliable Computer-Use BenchmarkintroducesOSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments