Microsoft has introduced a new AI model called Fara-7B, and it is designed to use your computer just like a human, but with a major advantage: it is tiny enough to run right on your device.
What Makes Fara-7B Different?
Forget the huge, cloud-hungry AI systems. Fara-7B is a Small Language Model (SLM) with only 7 billion parameters, making it incredibly compact and efficient. Microsoft calls it their first "agentic" SLM built for computer use.
-
- PC Control from a Screenshot: Instead of relying on massive cloud computing power, Fara-7B simply looks at a single screenshot of your screen and decides what to do next. It can then perform actions like clicking, typing, and navigating web pages.
- On-Device, Faster, and Private: Its small size means it can run directly on your PC. This is a huge win because it leads to:
- Reduced Lag (Latency): Actions happen instantly without waiting for the cloud.
- Improved Privacy: Your data stays local on your device.
- Affordable to Run: Microsoft estimates a complete task costs only about 2.5 cents, a fraction of the cost of using massive cloud models like GPT-4.
Microsoft in its official release describes the model as, “With only 7 billion parameters, Fara-7B achieves state-of-the-art performance within its size class and is competitive with larger, more resource-intensive agentic systems that depend on prompting multiple large models. Fara-7B’s small size now makes it possible to run CUA models directly on devices. This results in reduced latency and improved privacy, as user data remains local.”
An AI that Acts, Not Just Writes
Fara-7B is a Computer Use Agent (CUA). Unlike traditional AIs that just generate text replies, this model is built to act. It condenses the work that usually requires multiple large, complex AI systems into a single, straightforward model.
Trained to Act Human
To teach Fara-7B how to be a good assistant, Microsoft created a massive training system called FaraGen.
-
- This system simulated realistic, multi-step tasks on over 70,000 real websites, including human-like mistakes and retries.
- The system then filtered this data, resulting in over 1 million verified actions used to train the model.
High Performance for Its Size
Despite its compact design, Fara-7B's performance is impressive, holding its own against much larger, resource-heavy systems on various real-world web task benchmarks.
Available Now for Developers
Microsoft is making Fara-7B open-source and available on platforms like Hugging Face under an MIT license. They are also releasing a special version optimized for Copilot+ PCs running Windows 11, allowing users and developers to install and test the model locally. The goal is to make it easier for everyone to experiment with and advance technology for automating everyday web tasks.
To learn more, visit SecureITWorld!
Recommended For You:
AI Data Cloud Explained: The Role and Benefits of AI in Cloud Computing
