Reproduce “CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xinlei Yu, Gen Li, Ji Zhao, Zhangquan Chen, Congcong Wang, Shuai Dong, Zelin Peng, Yuqi Xu and 10 more

CUA-Master trains computer-use agents by turning real software tutorial workflows into executable tasks and using observed failures to guide later training. It composes reusable operations and retries failed tasks while preserving successful prefixes. Its 27-billion-parameter model improves over similarly sized UI-Mate-27B across four benchmarks and is competitive with larger systems, but the paper does not isolate which components drive gains.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.