# Microsoft's Fara1.5-27B targets computer-use agents

> A 27B-parameter vision-language model built to drive browsers and desktop apps like a human operator.

Published by The Open Weights on Jul 17, 2026. Canonical: https://theopenweights.com/news/fara1-5-27b-ni3f

## Key facts

- Company: Microsoft
- Model: Fara1.5-27B
- Version: 1.5
- Category: Vision-Language
- Modalities: Vision-Language
- License: Other (Open weights, commercial use allowed)
- Parameters: 27B
- Context window: 262K tokens
- Architecture: Qwen3_5ForConditionalGeneration
- Precision: BF16
- Weights size: 54.7 GB
- Significance: notable
- Published: 2026-07-17
- Last verified: 2026-08-06
- Hugging Face: https://huggingface.co/microsoft/Fara1.5-27B
- Canonical URL: https://theopenweights.com/news/fara1-5-27b-ni3f

Microsoft has published Fara1.5-27B, a vision-language model designed for computer-use automation, on [Hugging Face](https://huggingface.co/microsoft/Fara1.5-27B). Rather than answering questions from a chat box, the model is built to perceive graphical interfaces and take actions across web browsers and desktop applications.

At 27 billion parameters, Fara1.5-27B sits in the mid-size tier of open-weight releases—large enough to handle the visual reasoning that agentic tasks demand, but small enough to run without the infrastructure that frontier systems require. It is a dense model rather than a mixture-of-experts design, and it is distributed under a custom license.

## Why it matters

Computer-use agents are one of the more contested frontiers in applied AI right now. Systems that can read a screen, click buttons, fill forms, and navigate multi-step workflows promise to automate work that has resisted traditional scripting. Making that capability available as open weights lets researchers and developers study, fine-tune, and self-host the technology instead of depending on a closed API.

Key details from the release:

- Vision-language model tuned for browser and desktop automation
- 27B parameters, dense architecture
- Released under a custom ("other") license

As with any agent that can operate a machine on a user's behalf, the practical questions will center on reliability, safety, and how well the model generalizes beyond the benchmarks it was trained against. Microsoft's decision to ship Fara1.5-27B openly gives the community a chance to probe those limits directly. Full documentation and weights are available on the model's [Hugging Face page](https://huggingface.co/microsoft/Fara1.5-27B).

## Get the model

- [Hugging Face](https://huggingface.co/microsoft/Fara1.5-27B)

## Sources

- [microsoft/Fara1.5-27B](https://huggingface.co/microsoft/Fara1.5-27B) — Hugging Face, Jul 17, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Microsoft's Fara1.5-27B targets computer-use agents", The Open Weights, Jul 17, 2026, https://theopenweights.com/news/fara1-5-27b-ni3f