OSWorld A
AI benchmark · Agentic
- Built by
- University of Hong Kong Salesforce Carnegie Mellon University Waterloo
- Domains
- Agentic
- Tier
- A why
- Snapshot
What is OSWorld?
OSWorld measures how well multimodal agents complete 369 open-ended real-world computer tasks across desktop and web apps in real operating-system environments, using execution-based evaluation.
Computer use mostly
“we create a benchmark of 369 computer tasks involving real web and desktop apps in open domains, OS file I/O, and workflows spanning multiple applications.”
arxiv.org/abs/2404.07972 ↗
Who built OSWorld
Led by The University of Hong Kong (XLANG Lab), with Salesforce Research, Carnegie Mellon University and University of Waterloo.
Paper: arxiv.org/abs/2404.07972 ↗
Lineage
Benchmarks this one is built on and built from: versions, subsets, extensions and the indexes that combine them.
Derived benchmarks
ChainWorld extension
“ChainWorld measures how well computer use agents complete long-horizon desktop workloads built by chaining atomic OSWorld tasks, under single-turn and multi-turn protocols.”
OSWorld 2.0 later version
OSWorld 2.1 later version
OSWorld 2.1 Offline later version
OSWorld-MCP extension
OSWorld-Verified later version
OSWorld-Verified (XLANG extension
Included in
“OSWorld, τ²-bench, BrowseComp, Terminal-Bench 2.0, GDPval-AA, SWE-Pro”
huggingface.co/datasets/FINAL-Bench/ALL-Bench-Leaderboard ↗
Tier A
Widely used: a lab reports it, or 50 or more models are scored on it.
How much stands behind a tier: usage, who backs it, a confirmed line from its own site or paper, and a clear read on X. High with three of the four, medium with two, low otherwise.
- 45 models scored
- in curated RL directories (2 pts)
On X · well regarded, high buzz, from 11 posts about OSWorld in the last month
OSWorld variants are the default yardstick for computer-use agents, cited by Microsoft's Mage, Yutori, and Google; one post questions partial-score reporting.
How we got to tier A
- Adoption: 45 models scored. Tier B.
- On X: well regarded, high buzz. “OSWorld variants are the default yardstick for computer-use agents, cited by Microsoft's Mage, Yutori, and Google; one post questions partial-score reporting.” Moved up to A.
- In curated RL directories: 2 points, a floor of C.
- Confirmed: a line from its own site or paper places it.
A clear view on X moves a tier one step up or down; most entries don't move. How tiers work.
Where OSWorld sits
Its domain on the map, live. Hover a logo for its name; click to open it. Open it on the full map
Next: browse all benchmarks. Entry from the RL Research daily scrape of public sources, 2026-11-05 snapshot.