Fara-7B - Microsoft's open-source agentic model for small computers.

What is Fara-7B?

Fara-7B is an open-source, agent-based Small Language Model (SLM) from Microsoft, focused on computer usage. The model helps users complete tasks such as filling out forms, searching for information, or booking trips by visually perceiving web pages and interacting with interface elements like the mouse and keyboard. With only 7 billion parameters, the model achieves top-tier performance in its size category and can run directly on devices, reducing latency and improving privacy. Fara-7B is trained on synthetic data, offering efficient task execution while prioritizing security. The model is currently released as an experimental version to invite community exploration and feedback, driving technological advancement.

Main functions of Fara-7B

• Automated task executionIt helps users complete various computer tasks by manipulating interface elements such as the mouse and keyboard, such as filling out forms, searching for information, booking trips, and managing accounts.
• Visual perception and interactionThe model can directly perceive web page content and perform operations such as clicking, scrolling, and input by predicting coordinates. It relies entirely on screen visual information and does not require additional auxiliary information (such as an accessibility tree).
• User interaction and controlWhen performing a task, Fara-7B will pause at critical points (such as when user privacy or important decisions are involved) and wait for user confirmation or input to ensure that the user has complete control over the task.
• Privacy protection and securityAll operations are completed on the local device, user data is not transmitted externally, and privacy and security are ensured through logging and sandbox environment.
• High-efficiency task executionThrough optimized model architecture and training methods, Fara-7B can complete tasks in fewer steps, making it more efficient and less costly than other similar models.

Technical principles of Fara-7B

• Vision-based interactionThe model perceives web page content through screenshots, directly simulating the interaction between human users and computers, without relying on the internal structure of the web page (such as the DOM tree or accessibility tree).
• Synthetic data generationTo address the shortage of high-quality training data, Microsoft developed a synthetic data generation pipeline. This pipeline generates large-scale, multi-step task data from publicly available web pages and task hints, avoiding the high costs of manual annotation.
• Multi-agent system trainingDuring the training phase, Fara-7B uses a multi-agent system to solve synthetic tasks, generating demonstration data for supervised fine-tuning. The agents include task planners, web page operators, and user simulators, which work collaboratively to complete the tasks.
• Single-model distillationFara-7B distills the complexity of multi-agent systems into a single model, enabling it to operate independently and simplifying deployment and usage.
• Strengthen security mechanismsThe model incorporates safe data during training, enabling it to reject harmful tasks. Simultaneously, all operations are logged and auditable, ensuring complete user control over the model's behavior.

Fara-7B Project Address

• Project official website: https://www.microsoft.com/en-us/research/blog/fara-7b-an-efficient-agentic-model-for-computer-use/
• GitHub repositoryhttps://github.com/microsoft/fara
• HuggingFace model libraryhttps://huggingface.co/microsoft/Fara-7B
• Technical Papers: https://www.microsoft.com/en-us/research/wp-content/uploads/2025/11/Fara-7B-An-Efficient-Agentic-Model-for-Computer-Use.pdf

Application scenarios of Fara-7B

• Office AutomationThe Fara-7B can automatically process documents, compose emails, and enter data, significantly improving office efficiency.
• Information retrieval and organizationThe model can quickly search and summarize online information, helping users efficiently obtain the information they need.
• e-commerceThe Fara-7B can automatically search for products, compare prices, and assist with order placement, optimizing the shopping experience.
• Travel planningThe model can intelligently arrange itineraries, book flights and hotels, and simplify the travel preparation process.
• Online learningThe Fara-7B can automatically search for courses and organize learning materials to help users learn efficiently.