Microsoft Unveils Fara-7B: A Lightweight Language Model with GPT-4o-Level Performance, Running Directly on PCs
2025-11-25 / Read about 0 minute
Author:小编   

On November 24 (local time), Microsoft unveiled its inaugural agent-based compact language model, Fara-7B, tailored explicitly for computer operations. This model, encompassing 7 billion parameters, is classified as a Computer Use Agent (CUA). It empowers users by autonomously executing tasks via interactive devices like mice and keyboards, operating seamlessly on end-user devices without the need for external servers. Fara-7B stands out for its lightweight architecture and rapid response capabilities, setting a new benchmark for models of its scale. By circumventing the dependency on resource-intensive cloud-based models, it achieves minimal latency and bolsters data privacy, particularly on systems with constrained resources. Presently, the model has been made openly accessible on Microsoft Foundry and Hugging Face platforms and has been seamlessly incorporated into the Magentic-UI research prototype.