
Cowork-Gym constructs executable work environments, validates and repairs them through a build–execute–revise loop, and collects trajectories with a stateful user simulator. The resulting environments support supervised fine-tuning, reinforcement learning, and on-policy distillation.
Cowork-Gym 构建可执行的工作环境,通过构建、执行与修订闭环进行验证和修复,并借助有状态用户模拟器采集交互轨迹,支持监督微调、强化学习与在线策略蒸馏。
Anni Zou, Cheng Chen, Jiaji Deng, Weikang Zhou, Xinji Mai, Yue Wu, Jianing Lin, Lipeng Xie, Lixiong Qin, Qingyue Zhao, Que Shen, Shuchang Tao, Xuchen Pan, Yaliang Li, Yunpeng Zhai, Yushuo Chen, Zhihai Wang, Zhuo Zhang, Zhaoyang Liu, Bolin Ding, Jingren Zhou. "Cowork-Gym: Synthesizing Interactive Work Environments at Scale." arXiv:2610.17670 (2026).