OpenCUA: Open Foundations for Computer-Use Agents

Diyi Yang (Stanford University) · Tianbao Xie (the University of Hong Kong, University of Hong Kong) · Jiaqi Deng (The University of Hong Kong) · Xiaochuan Li (Tsinghua University) · Junlin Yang (Tsinghua University) · Jixuan Chen (University of California, San Diego) · Xinyuan Wang (University of Hong Kong) · Yiheng Xu (University of Hong Kong) · Junli Wang (Alibaba Group) · Tao Yu (University of Hong Kong) · Bowen Wang (University of Hong Kong) · Dunjie Lu (SUN YAT-SEN UNIVERSITY) · Xiaole Guo (University of Hong Kong) · Chen Wu (Carnegie Mellon University) · Zhennan Shen (Shanghai Jiaotong University) · Zhuokai Li (University of Hong Kong) · Ryan Li (Computer Science Department, Stanford University) · Junda Chen (Harbin Institute of Technology) · Boyuan Zheng (The University of Hong Kong) · LI PEIHANG (University of Hong Kong) · Fangyu Lei (Institute of automation, Chinese academy of science, Chinese Academy of Sciences) · Ruisheng Cao (Shanghai Jiaotong University) · Yeqiao Fu (University of Hong Kong) · Dongchan Shin (University of Hong Kong) · Martin Shin (University of Hong Kong) · Hu Jiarui (University of Hong Kong) · Yuyan Wang (Johns Hopkins University) · Yuxiao Ye (The Hong Kong University of Science and Technology) · Danyang Zhang (Shanghai Jiao Tong University) · Yipu Wang (Institute of automation, Chinese academy of science, Chinese Academy of Sciences) · Heng Wang (University of Illinois Urbana-Champaign) · Victor Zhong (University of Waterloo) · Y.Charles (Moonshot AI) · Zhilin Yang (Tsinghua University, Tsinghua University)

agentnetannotation infrastructurechain-of-thought reasoningcomputer-use agentscua benchmarksgeneralization large-scale datasetopen foundationsopen-source frameworkopencua-72bperformance gains state-action pairs test-time computation vision-language models

Vision-language models have demonstrated impressive capabilities as computer-use agents (CUAs) capable of automating diverse computer tasks. As their commercial potential grows, critical details of the most capable CUA systems remain closed. As these agents will increasingly mediate digital interactions and execute consequential decisions on our behalf, the research community needs access to open CUA frameworks to study their capabilities, limitations, and risks. To bridge this gap, we propose OpenCUA, a comprehensive open-source framework for scaling CUA data and foundation models. Our framework consists of: (1) an annotation infrastructure that seamlessly captures human computer-use demonstrations; (2) AgentNet, the first large-scale computer-use task dataset spanning 3 operating systems and 200+ applications and websites; (3) a scalable pipeline that transforms demonstrations into state–action pairs with reflective long Chain-of-Thought reasoning that sustain robust performance gains as data scales. Our end-to-end agent models demonstrate strong performance across CUA benchmarks. In particular, OpenCUA-72B achieves an average success rate of 45.0% on OSWorld‑Verified, establishing a new state-of-the-art (SOTA) among open-source models. Further analysis confirms that our approach generalizes well across domains and benefits significantly from increased test-time computation. We release our annotation tool, datasets, code, and models to build open foundations for further CUA research.