# Microsoft Releases Fara-7B Vision Agent Model

> The 7-billion-parameter model is designed to understand and interact with graphical user interfaces, building on Alibaba's open-source Qwen2.5-VL.

Published by The Open Weights on Oct 30, 2025. Canonical: https://theopenweights.com/news/fara-7b-vccl

## Key facts

- Company: Microsoft
- Model: Fara-7B
- Version: v1
- Category: Vision-Language
- Modalities: Vision-Language
- License: MIT (Open weights, commercial use allowed)
- Parameters: 7B
- Context window: 128K tokens
- Architecture: Qwen2_5_VLForConditionalGeneration
- Precision: BF16
- Weights size: 16.6 GB
- Significance: notable
- Published: 2025-10-30
- Last verified: 2026-06-17
- Hugging Face: https://huggingface.co/microsoft/Fara-7B
- Canonical URL: https://theopenweights.com/news/fara-7b-vccl

Microsoft has introduced Fara-7B, a new 7-billion-parameter vision-language model aimed at a specific and challenging task: controlling a computer. Unlike general-purpose multimodal models, Fara-7B is designed to function as an agent, interpreting graphical user interfaces (GUIs) to understand and execute tasks.

This specialization allows the model to go beyond simply describing what's on a screen. The goal is for Fara-7B to comprehend the layout, elements, and interactive possibilities within an application, paving the way for more sophisticated AI-powered automation and assistance.

Interestingly, Fara-7B is not built from the ground up. According to its [official model card](https://huggingface.co/microsoft/Fara-7B), the model is based on Alibaba's recently released Qwen2.5-VL. This approach highlights a growing trend of major AI labs building upon and refining foundational models released by others, accelerating the pace of innovation across the open-source community.

## Why it matters

The release of specialized agent models like Fara-7B under a permissive MIT license provides a powerful building block for developers. It opens up new possibilities for creating advanced accessibility tools, automating repetitive software tasks, and developing more capable personal AI assistants that can interact with technology the same way humans do: by seeing and clicking.

## Get the model

- [Hugging Face](https://huggingface.co/microsoft/Fara-7B)

## Sources

- [microsoft/Fara-7B](https://huggingface.co/microsoft/Fara-7B) — Hugging Face, Oct 30, 2025

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Microsoft Releases Fara-7B Vision Agent Model", The Open Weights, Oct 30, 2025, https://theopenweights.com/news/fara-7b-vccl