Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses
Computer Science > Computation and Language arXiv:2610.40747 (cs) [Submitted on 29 Oct 2026] 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.
计算机科学 > 计算与语言 arXiv:2610.40747 (cs) [提交于 2026 年 10 月 29 日] 标题：Mage-VL-Cowork：一种从进化工具链中学习混合计算机使用方法的完整训练方案。作者：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.
摘要：有效的计算机使用智能体（CUA）应在单一策略内无缝结合视觉图形用户界面（GUI）交互与代码及终端执行。我们提出了 Mage-VL-Cowork，这是一个 4B 参数的混合型 CUA，它通过端到端、渐进式的五个阶段课程（涵盖中期训练、SFT 和 RL）从一个通用的多模态对齐检查点训练而来。
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%.
为了大规模支持这一课程，Mage-CUA-Forge 用可重用的、基于技能条件的生成工作流和特定于应用程序的准备器取代了针对每个任务的智能体构建，合成了涵盖 6,915 种技能的 293,806 个桌面和 Web 任务，同时将每个任务的重复性开销降低了 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.
我们没有直接蒸馏默认的教师模型，而是进化了教师工具链并实例化了专门的 CUA 工具链，以在特定阶段监督的同时产生高质量的轨迹。Mage-VL-Cowork-4B 通过不同的训练阶段，最终在 OSWorld-Verified 上达到了 66.0% 的准确率，且推理成本更低，在同等参数规模下实现了最先进的性能，并推动了端侧智能的前沿发展。