Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses
Paper · https://arxiv.org/abs/2610.40747
arXiv:2610.40747v1 Announce Type: new 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.
First seen by Oriel on .
AI summary
Mage-VL-Cowork is a 4B hybrid computer-use agent trained through a five-stage curriculum, reaching 66.0% on OSWorld-Verified with efficient reasoning costs.
- The policy combines visual GUI interaction with code and terminal execution within a single model.
- Mage-CUA-Forge synthesized 293,806 desktop and web tasks spanning 6,915 skills using reusable, skill-conditioned workflows and application-specific preparers.
- The system evolves teacher harnesses and creates specialized CUA harnesses to generate trajectories and stage-specific supervision instead of directly distilling a default teacher.
- The curriculum covers midtraining, supervised fine-tuning, and reinforcement learning, while reducing recurring per-task expenditure by 82.6%.
- Mage-VL-Cowork-4B achieves state-of-the-art performance at its parameter size and is designed to advance on-device intelligence.
#computer-use agents #multimodal models #reinforcement learning #on-device intelligence
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2610.40747v1 | https://arxiv.org/abs/2610.40747 Read discussion |