microsoft/Mage-VL-Cowork
AI model · https://huggingface.co/microsoft/Mage-VL-Cowork
First seen by Oriel on .
AI summary
The Mage-VL-Cowork model card presents a 4B vision-language agent and five-stage training recipe for unified GUI and terminal computer use, reporting 66.0% on OSWorld-Verified.
- The policy dynamically routes visual layouts and live application state through computer_use, while using persistent bash for structured edits and deterministic verification, without an external orchestrator.
- Mage-CUA-Forge generated 293,806 tasks across 6,915 atomic skills and reduced recurring synthesis cost by 82.6%, from $0.906 to $0.158 per task.
- The curriculum progresses from GUI grounding and standalone coding to Hybrid SFT and Online Hybrid RL, improving OSWorld-Verified results from 45.56% to 66.04%.
- Online Hybrid RL increased accuracy by 10.16 percentage points over Hybrid SFT while reducing model calls by 13.0% and output tokens by 25.4%.
- The model uses a 32,768-token active window, retaining three recent screenshots and four complete responses while folding earlier interaction history into summaries.
#computer use #vision-language models #reinforcement learning #gui automation
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 207, latest observation ; +103 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 3,080, latest observation ; +2,328 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | microsoft/Mage-VL-Cowork | https://huggingface.co/microsoft/Mage-VL-Cowork Read discussion |