New 4B Hybrid Agent Achieves State-of-the-Art Performance in Computer Use Tasks
In short: A new research paper introduces Mage-VL-Cowork, a 4B parameter computer-use agent that combines GUI interaction with code and terminal execution. By utilizing a new synthetic data framework called Mage-CUA-Forge, the developers reduced per-task training expenditure by 82.6% while reaching 66% on the OSWorld-Verified benchmark.
This summary was generated automatically by AI from arXiv cs.CL's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1Introduction of Mage-VL-Cowork-4B for hybrid GUI/terminal task execution.
- 2Deployment of Mage-CUA-Forge to synthesize nearly 300,000 tasks across 6,915 skills.
- 3Reduction of per-task training costs by 82.6%.
| Parameter | Before | Now |
|---|---|---|
| OSWorld-Verified benchmark score | Not specified | 66.0% |
Why it matters
Efficient, small-scale computer-use agents are essential for automating complex workflows that span multiple interfaces. This approach demonstrates that high-performance agents can be trained much more cheaply than previously thought, lowering the barrier for building specialized automation tools.
What it means for AI agents and contact centers
If you are building autonomous agents for contact-center automation or desktop assistance, this model offers a path to lower latency and local execution due to its 4B size. Test this model against your existing workflows that require switching between GUI and terminal commands to assess its effectiveness compared to larger, more expensive foundation models.
🧪 Worth Testing
Testing is recommended to evaluate if this compact model can handle specialized agentic tasks with higher efficiency and lower operational overhead.
Sources
- arXiv cs.CLOfficialPrimary sourceOriginal article →„Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses“2 Nov 2026, 07:00Licence: arXiv metadata CC0; papers under their own licences · our summary (content changed)
- Published by source
- 2 Nov 2026, 07:00
- Found by our system
- 2 Nov 2026, 07:22
- Summary generated
- 2 Nov 2026, 07:22
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy