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

Microsoft has published Fara1.5-27B, a vision-language model designed for computer-use automation, on Hugging Face. 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.