Reproduce “CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents” 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
Zhongwei Xie, Yiming Du, Jiaxin Bai, Baixuan Xu, Chunyang Li, Xiyu Ren, Wei Fan, Tianshi Zheng and 5 more
CoworkEnv constructs coworker tasks by linking requests, records, tool interfaces, and grading checks across services. It uses teacher demonstrations for supervised fine-tuning and student interactions for reinforcement learning. For Qwen3-VL-8B-Instruct, supervised fine-tuning raises the Claw-Eval mean from 46.6 to 64.1, with the largest gain on multi-turn tasks. These results do not isolate the contribution of CoworkEnv’s construction methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.