# Microsoft's Mage-VL-Cowork takes on computer-use tasks

> A new vision-language agent model aims to operate GUIs, write code, and work the terminal.

Published by The Open Weights on Oct 30, 2026. Canonical: https://theopenweights.com/news/mage-vl-cowork-509j

## Key facts

- Company: Microsoft
- Model: Mage-VL-Cowork
- Version: Cowork
- Category: Vision-Language
- Modalities: Vision-Language
- License: Other (Open weights, commercial use allowed)
- Context window: 262K tokens
- Architecture: llava_onevision2
- Precision: BF16
- Weights size: 9.5 GB
- Significance: notable
- Published: 2026-10-30
- Last verified: 2026-11-02
- Hugging Face: https://huggingface.co/microsoft/Mage-VL-Cowork
- Canonical URL: https://theopenweights.com/news/mage-vl-cowork-509j

Microsoft has published **Mage-VL-Cowork**, a vision-language model built to act as a computer-use agent, on [Hugging Face](https://huggingface.co/microsoft/Mage-VL-Cowork). Rather than simply describing images, the model is designed to take actions — navigating graphical interfaces, writing and running code, and operating a terminal.

The release reflects a broader shift in open-weight AI toward agentic systems that can perceive a screen and manipulate it directly. Where earlier vision-language models focused on captioning and visual question answering, this generation of "computer-use" models treats the desktop itself as the environment to be understood and controlled.

## What it targets

- GUI interaction, including clicking through and navigating on-screen interfaces
- Coding tasks that combine visual context with code generation
- Terminal and command-line workflows

Microsoft has not published parameter counts, context length, or benchmark figures alongside the initial listing, and the model carries a non-standard "other" license, so teams evaluating it for production should review the terms on the model card carefully.

Why it matters: computer-use agents are one of the most contested frontiers in applied AI, with closed offerings from several major labs. An openly available model aimed at the same GUI-and-terminal territory gives researchers and developers a base to study, fine-tune, and benchmark against — assuming the licensing fits their use. More details are on the [Hugging Face repository](https://huggingface.co/microsoft/Mage-VL-Cowork).

## Get the model

- [Hugging Face](https://huggingface.co/microsoft/Mage-VL-Cowork)

## Sources

- [microsoft/Mage-VL-Cowork](https://huggingface.co/microsoft/Mage-VL-Cowork) — Hugging Face, Oct 30, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Microsoft's Mage-VL-Cowork takes on computer-use tasks", The Open Weights, Oct 30, 2026, https://theopenweights.com/news/mage-vl-cowork-509j