Microsoft releases Mage-VL-Cowork 4B parameter model
Original title: microsoft/Mage-VL-Cowork — new model trending #30 on Hugging Face
Mage-VL-Cowork is a 4B parameter hybrid computer use model achieving 66.0% on OSWorld-Verified.
Microsoft's Mage-VL-Cowork is a single vision-language policy handling GUI interaction and terminal execution. Trained in five stages from a 4B checkpoint using evolved teacher harnesses and workflow-first synthesis, it reaches 66.0% on OSWorld-Verified. Online RL adds +10.16 pp and reduces tokens by 25.4%.
- Achieves 66.0% on OSWorld-Verified at 4B parameters
- Combines computer_use and bash channels
- Uses five-stage specialize-then-unify curriculum
- Online RL adds +10.16 pp and reduces tokens
- Trained on 293,806 tasks across 6,915 skills
- Organizations
- Microsoft
- AI models
- Mage-VL-Cowork
From the original article
13 min read at huggingface.coimage-text-to-text · 1,541 downloads · 156 likes · license apache-2. 0
By microsoft · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.