AI Policy Wiki
Dashboard
⚲

XLANG Lab (University of Hong Kong)

medium confidence · updated 2026-09-22

Research group at the University of Hong Kong led by Tao Yu that built and maintains the OSWorld computer-use benchmark (2024) and its long-horizon successor OSWorld 2.0 (2026).

The XLANG Lab is a research group at the University of Hong Kong whose corresponding author on its benchmark papers is Tao Yu. It is the maintaining organisation of OSWorld, the computer-use benchmark in which an agent operates a real Ubuntu desktop through screenshots, mouse and keyboard.

Activities

The lab's April 2024 paper introduced OSWorld as a real-computer environment with 369 execution-scored tasks, written by nine student authors over roughly 1,800 person-hours with co-authors at Carnegie Mellon University, Salesforce Research and the University of Waterloo; it reported 72.36% human success against 12.24% for the best model (OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments (Xie et al., 2024)). In June 2026 the lab, credited as "XLANG Lab and Collaborators" with project co-leads Mengqi Yuan, Zilong Zhou and Xinzhuang Xiong, released OSWorld 2.0, 108 long-horizon workflows across 31 self-hosted web services and professional desktop applications, with collaborators at UC San Diego, Columbia, UC Santa Barbara, Mila, Snorkel AI, the University of Wisconsin–Madison, Alibaba Qwen, Ohio State University and others; the paper acknowledges Snorkel AI as research and data partner and support from a Google Research gift fund (OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks (Yuan et al., 2026)). Both papers release their environments, tasks and evaluation code publicly.

Relationships