Fara-7B - Microsoft's open-source computer-operated Agent assistant model

What is Fara-7B?

Fara-7B is a 7 billion parameter scale Computer Operating Agent (CUA) model released by Microsoft open source, based on the Qwen2.5-VL-7B architecture. By visually parsing web page screenshots, it performs clicks, inputs, and other operations on the screen without relying on additional accessibility trees or collaboration of multiple large models, and can be run directly on Windows 11 locally with support for NPU acceleration for lower latency and better privacy protection.Fara-7B has been shown in public benchmarks such as WebVoyager, Online-Mind2Web, and others to Excellent performance with high task success rate and some tasks ahead of peer models. Adopts a new synthetic data generation process for training, containing a large number of task trajectories and auxiliary task data, with a focus on supervised fine-tuning.

Functional features of Fara-7B

• vision-driven operation: Perform clicking, typing, scrolling, etc. directly on the screen by visually parsing screenshots of web pages, without relying on additional accessibility trees or collaboration of multiple large models.
• Local operation and privacy protection: Runs natively on Windows 11 and supports NPU acceleration for low latency and better privacy protection.
• Well-established security mechanisms: Stop at "critical points" to ask for user consent for sensitive operations, all operations are logged, need to be run in a sandbox environment, and examples of refusing to perform inappropriate tasks are included in the training.
• High performance: Excellent performance in public benchmarks such as WebVoyager, Online-Mind2Web, etc., with high task success rate and some tasks leading the peer models.
• Open Source and Ease of Use: Released and open-sourced under the MIT License at Microsoft Foundry and Hugging Face, and integrated into the Magentic-UI research prototype, providing quantized and optimized versions for ease of use and development.

Core Benefits of Fara-7B

• Vision-driven and direct manipulation: Fara-7B performs operations directly on the screen by visually parsing screenshots of web pages, eliminating the need to rely on complex accessibility trees or collaboration of multiple models, making operations more intuitive and efficient.
• Local operation and privacy protection: Supports running natively on Windows 11, combined with NPU acceleration for low-latency response while ensuring user data privacy.
• Strong security mechanismsThe user's consent is sought at "critical points" when performing sensitive operations. All operations are recorded and run in a sandboxed environment, effectively preventing inappropriate operations.
• High performance and success rate: Excellent performance in a number of public benchmarks, with a high task success rate and some tasks ahead of peer models, demonstrating outstanding performance.

What is the official website of Fara-7B

• Project website:: https://www.microsoft.com/en-us/research/blog/fara-7b-an-efficient-agentic-model-for-computer-use/
• GitHub repository:: https://github.com/microsoft/fara
• HuggingFace Model Library:: https://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

People for whom Fara-7B is intended

• Developers and Researchers: The open source nature of Fara-7B becomes an ideal tool for developers and researchers to use its powerful features for secondary development, model optimization, and algorithm research.
• automated task manager: For users who need to perform complex automation tasks on a computer, such as data entry and web page automation operations, the Fara-7B enables efficient operation with visual drive.
• Privacy and Security Needs: Users focused on data privacy and security will benefit from Fara-7B's locally run and sandboxed environment designed to ensure the security of sensitive information.
• Users with high performance requirements: In scenarios that require fast response and low latency, such as real-time web interactions and automated testing, the Fara-7B's high-performance performance is able to meet the demand.
• For new technology explorers: Users interested in the latest AI technologies can explore the combination of computer vision and natural language processing and its potential for practical applications with the Fara-7B.