arXiv cs.CL· 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·· 11 小时前AI 评分49
Mage-VL-Cowork：从进化式 harness 学习混合计算机使用的完整训练配方
Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses
AI 导读
Mage-VL-Cowork 是一个 4B 混合计算机使用智能体（CUA），从通用多模态对齐 checkpoint 出发，经 midtraining、SFT、RL 五阶段渐进课程端到端训练，在 OSWorld-Verified 上达到 66.0%，在同参数规模下取得 SOTA。
正文 · 原文
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
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.
| Comments: | Project page: this https URL |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.40747 [cs.CL] |
| (or arXiv:2610.40747v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.40747 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xiaoyi Zhang [view email]
[v1]
Thu, 29 Oct 2026 15:54:48 UTC (1,044 KB)
来源：arXiv cs.CL · arxiv.org