Mage-VL-Cowork

Use it to carry out multi-step desktop and web tasks, including visual interactions and programmatic file or system operations.

Availability

A single unified checkpoint, microsoft/Mage-VL-Cowork, is listed on Hugging Face.

Pricing model

Not established

Deployment

The text describes on-device intelligence; checkpoint is available as safetensors.

Privacy

Not established

Licence

Apache-2.0

• The 4B policy combines screenshot-grounded GUI interaction and persistent code/terminal execution. ^[1]

  Trained end-to-end from an intermediate **4B [`Mage-VL`](https://huggingface.co/microsoft/Mage-VL)** checkpoint through a progressive **five-stage specialize-then-unify curriculum**, it combines screenshot-grounded GUI interaction (`computer_use`) and persistent code/terminal execution (`bash`) inside a **single vision-language policy** — with no external orchestrator.

• The model dynamically selects between computer_use and bash without an external multi-agent router. ^[1]

  - **Internalized turn-level channel routing.** Without an external multi-agent router, the policy dynamically selects `computer_use` for visual layouts, canvas editing, and live in-memory application state, and switches to `bash` for structured programmatic edits (spreadsheets, documents, presentations) and deterministic ground-truth verification.

• The model is available as a unified checkpoint at microsoft/Mage-VL-Cowork. ^[1]

  - **Specialize-then-unify five-stage curriculum (+20.44 pp).** Starting from an intermediate 4B `Mage-VL` checkpoint, the model builds standalone visual grounding (**30.6% → 56.9%** on OSWorld-G; **78.8% → 87.2%** on ScreenSpot-v2), GUI execution, and coding/terminal competence before unifying them via Hybrid SFT and asynchronous Online Hybrid RL (**45.56% → 55.84% → 66.0%**).

• The text describes the model as advancing on-device intelligence. ^[1]

  - **Best-in-class 4B hybrid computer use (66.0% OSWorld-Verified).** Achieving the highest score at the 4B scale, `Mage-VL-Cowork` proves that an efficient small VLM can master end-to-end computer use — leading on **Operating System tasks at 87.5%**, reaching **62.7% on Multi-Application Workflows**, and advancing the frontier of **on-device intelligence**.

• The model reaches 66.0% on OSWorld-Verified under a 50-step budget. ^[1]

  At **4B parameters**, **Mage-VL-Cowork** reaches **66.0% on OSWorld-Verified** under a 50-step budget — achieving the **best performance among 4B models** and demonstrating that an **efficient small VLM can excel at complex computer use**, pushing the frontier of **on-device intelligence**.

• The checkpoint is identified as Apache-2.0 licensed. ^[1]

  - **Specialize-then-unify five-stage curriculum (+20.44 pp).** Starting from an intermediate 4B `Mage-VL` checkpoint, the model builds standalone visual grounding (**30.6% → 56.9%** on OSWorld-G; **78.8% → 87.2%** on ScreenSpot-v2), GUI execution, and coding/terminal competence before unifying them via Hybrid SFT and asynchronous Online Hybrid RL (**45.56% → 55.84% → 66.0%**).