| title | Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses |
|---|---|
| authors | Xiaoyi Zhang, Haoqing Wang, Xinjie Zhang, Xiang Ming, Xing Wang, Senqiao Yang, Kaichen Zhang, Zhaoyang Jia, Andong Tan, Xinjiang Wang, Jinghao Guo, Yifei Shen, Xiao Li, Peng Zhang, Xun Guo, Jiahao Li, Shicheng Zheng, Jinglu Wang, Zongyu Guo, Wenxuan Xie, Bin Li, Yan Lu |
| date | 2026-10-29 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.40747 |
| affiliation | Microsoft |
| upvotes | 12 (Hugging Face Daily Papers, 2026-11-05) |
| links | abs pdf hf |
| cite | |
| abstract | Effective computer-use agents (CUAs) should seamlessly combine visual GUI interaction with code and terminal execution within a single policy. We present Mage-VL-Cowork, a 4B hybrid CUA trained from a general multimodally aligned checkpoint through an end-to-end, progressive five-stage curriculum across midtraining, SFT, and RL. To power this curriculum at scale, Mage-CUA-Forge replaces per-task agent construction with reusable, skill-conditioned generation workflows and application-specific preparers, synthesizing 293,806 desktop and web tasks across 6,915 skills while reducing recurring per-task expenditure by 82.6%. Rather than directly distilling a default teacher, we evolve the teacher harness and instantiate specialized CUA harnesses to yield high-quality trajectories alongside stage-specific supervision. Mage-VL-Cowork-4B finally climbs to 66.0% on OSWorld-Verified through different training stages with more efficient reasoning cost, achieving state-of-the-art performance at its parameter size and pushing forward the frontier of on-device intelligence. |
bibtex
@misc{zhang2026mage,
title={Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses},
author={Xiaoyi Zhang and Haoqing Wang and Xinjie Zhang and Xiang Ming and Xing Wang and Senqiao Yang and Kaichen Zhang and Zhaoyang Jia and Andong Tan and Xinjiang Wang and Jinghao Guo and Yifei Shen and Xiao Li and Peng Zhang and Xun Guo and Jiahao Li and Shicheng Zheng and Jinglu Wang and Zongyu Guo and Wenxuan Xie and Bin Li and Yan Lu},
year={2026},
eprint={2610.40747},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.40747}
}