CoworkEnv synthesizes training environments for long-horizon coworker agents
Original title: CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents
SFT on Qwen3-VL-8B-Instruct raised the Claw-Eval mean from 46.6 to 64.1, including a 31.9-point multi-turn gain.
CoworkEnv builds executable training environments for AI coworkers from workflow requirements, using programmatic validation and feedback repair. Teacher executions supply demonstrations for supervised fine-tuning, while student interactions supply rollouts and rewards for reinforcement learning. Compact PocketWorker models are trained from those resources, and an SFT-only extension is tested on ClawMark.
Unconfirmed: Support for policy optimization is described only as preliminary evidence from improving rewards.
- Qwen3-VL-8B-Instruct rose from 46.6 to 64.1 mean score on Claw-Eval after SFT.
- Multi-turn tasks accounted for a 31.9-point gain.
- Authors compare RL setups that vary reward processing and credit assignment.
- Improving training rewards are called preliminary evidence for policy optimization.
- AI models
- Qwen3-VL-8B-InstructPocketWorker
From the original article
1 min read at huggingface.co
AI coworkers must coordinate tools, communicate across turns, and act on changing workspace state while respecting workflow constraints. Constructing training experience for these capabilities requires coherent requests, service records, tool interfaces, and grading checks. We present CoworkEnv, a capability-guided framework for synthesizing executable training environments for AI coworkers.
By Zhongwei Xie, Yiming Du, Jiaxin Bai · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.