Reproduce “Cowork-Gym: Synthesizing Interactive Work Environments at Scale” 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
Anni Zou, Cheng Chen, Jiaji Deng, Weikang Zhou, Xinji Mai, Yue Wu, Jianing Lin, Lipeng Xie and 13 more
Cowork-Gym generates environments for agents working with people through shared files and tools. It builds tasks, workspaces, user policies, and hybrid checks, then runs and revises environments; simulated users respond to progress and can change inputs. Qwen models improve in most comparisons; Ours-27B scores 0.8400 on PawBench under OpenClaw, but WildClawBench is a weakness and the pool is not representative.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.