Microsoft introduces a new open-source small language model, Fara-7B.

Microsoft has released Fara-7B, its first open-source intelligent agent model designed specifically for PC operation. Composed of 7 billion parameters, it supports Windows 11 and boasts class-leading performance. The model performs actions such as clicking and inputting by visually analyzing webpage screenshots, without relying on additional accessibility trees or collaboration between multiple large models. Microsoft has built a new synthetic data generation process to simulate webpage tasks. Fara-7B is based on the Qwen2.5-VL-7B platform and supports 128k context.