What capabilities does OSWorld-Verified test?

OSWorld-Verified evaluates computer use, planning, tool calling.

What is the current top score on OSWorld-Verified?

The top reported score is 85.0% by Claude Fable 5, across 16 models reporting (9 from frontier labs).

OSWorld-Verified is available under Apache-2.0.

Frontier

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

Publisher: XLANG Lab
Capabilities: Computer Use Planning Tool Calling
Domain: agentic
Format: Custom
Size: 361 tasks
License: Apache-2.0
Published: Oct 2023
Notable for: Benchmark for evaluating computer use, planning and tool calling in the agentic domain.
Canonical: os-world.github.io
Also on: github.com/xlang-ai/OSWorld

Cite

Notes

Only stored in your browser.

Attribution

Top score 85.0% by Claude Fable 5 - 16 models reporting (9 frontier)

OSWorld-VerifiedBar chart with 16 bars. Highest value: Claude Mythos Preview at 85.4.

16 models

blog · 2025

XLANG Lab's audited and bug-fixed re-release of OSWorld, removing flaky / impossible / underspecified tasks for a cleaner computer-use leaderboard.

blog · 2025

XLANG Lab's audited and bug-fixed re-release of OSWorld, removing flaky / impossible / underspecified tasks for a cleaner computer-use leaderboard.

What is OSWorld-Verified?: Cleaned, human-validated subset of OSWorld tasks designed for stable cross-lab comparison of computer-use agents.
What capabilities does OSWorld-Verified test?: OSWorld-Verified evaluates computer use, planning, tool calling.
What is the current top score on OSWorld-Verified?: The top reported score is 85.0% by Claude Fable 5, across 16 models reporting (9 from frontier labs).
What license is OSWorld-Verified under?: OSWorld-Verified is available under Apache-2.0.