Microsoft Launches Fara-7B, an Experimental Model Designed to Perform Tasks for Users

At the recent Microsoft Ignite conference, one of the key areas of focus when it came to the next stage of agentic AI was on computer use and how agents can perform tasks on behalf of users. Now, Microsoft has announced a new small language model (SLM), Fara-7B, specifically designed for this purpose.

The launch of Fara-7B builds on Microsoft’s earlier Phi SLM, which debuted in 2024. However, Fara-7B comes with significantly enhanced specifications and is designated as a Computer Use Agent (CUA) model.

AI Agent & Copilot Summit is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents. Building on its 2025 success, the 2026 event takes place March 17-19 in San Diego. Get more details.

What is Fara-7B?

Fara-7B is an open-weight model that, according to Microsoft, lowers “the barrier to experimenting with and improving CUA technology.” It goes beyond simple text generation and actively interacts with computer interfaces to complete web-based tasks on behalf of users.

These tasks could include making reservations, searching for information, or shopping for products. Fara-7B consists of seven billion parameters — to put that in context, Meta’s Llama Large Language Model boasts 1.4 trillion — meaning it can be run directly on devices. This enhances privacy while reducing latency.

The model works by visually “seeing” a webpage and does not require a separate model to interpret a user’s screen. It essentially operates in the same way a user would when scrolling through the internet.

Microsoft has released Fara-7B in its experimental phase, actively seeking feedback through user experimentation. Users are encouraged to test the model in a sandboxed environment to explore its computer use capabilities, such as document filing and web account management, while avoiding high-risk data or sensitive domains.

Currently, users can access Fara-7B through Microsoft Foundry and Hugging Face, as well as with Magentic-UI, Microsoft’s open-source research prototype of a human-centered agent. Additionally, Fara-7B can be downloaded on Copilot+ PCs.

Closing Thoughts

Regarding Fara-7B, Microsoft researchers have said:

Our current release is an experimental CUA model that achieves state-of-the-art results for its size, purely using supervised fine-tuning. We believe even stronger CUA models capable of running on-device are possible through improved multimodal base models and through Reinforcement Learning on live and sandboxed environments. These early days are about learning from the community and driving real-world experimentation to shape what comes next. 

Moving towards computer use and away from chatbots is a significant step for any company in the AI space. It works under the assumption that users are ready to hand over control to agents. However, Microsoft’s approach to this research is encouraging.

Deploying a model in its experimental phase demonstrates the company’s confidence in both its safety and capabilities. Doing so while explicitly inviting the Microsoft community to experiment is even bolder.

By asking the community to provide substantial feedback and engage in experimentation during the early stages of this technology, Microsoft adopts a powerful strategy. Not only will this help Microsoft understand the use cases where a computer use model like Fara-7B is best suited and desired by users, but it also involves users in the development process of the product.

Microsoft, like all the AI pioneers in the field, needs not only user support but also user input. This is especially important at a time when the focus is shifting from two-way conversations — human-agent — to a period where human commands are interpreted and acted upon by multiple agents — human-agent-agent-agent.

---