Archive · 28 Nov 2025
Microsoft Unveils Fara-7B, a Lightweight AI Model
Microsoft has introduced a new AI model called Fara-7B.
This model is special because it is small and can be run on a regular computer, unlike other AI models that need powerful cloud servers.
Fara-7B can look at a screenshot and make decisions, such as typing, clicking, and navigating the web.
It was trained using a system called FaraGen, which created realistic tasks on websites to teach the model.
The model is available for developers to use and is designed to make everyday computer tasks easier and more affordable.
It performs well on various benchmarks and is more cost-effective than larger AI models.
Microsoft unveiled Fara-7B, a lightweight AI model with 7 billion parameters.
Fara-7B is designed to run on devices, reducing latency and improving privacy.
The model uses a synthetic data pipeline called FaraGen for training.
Fara-7B performs well on benchmarks like Web Voyager and WebTailBench.
The model is available on Microsoft Foundry, Hugging Face, and Copilot+ PCs running Windows 11.
- Who
- Microsoft
- What
- Unveiled Fara-7B, a lightweight AI model
- When
- Not specified
- Where
- Not specified
- Why
- To create a compact AI model that can be run on devices for everyday computer tasks
- Model Name
- Fara-7B
- Parameters
- 7 billion
- Performance Benchmarks
- Web Voyager: 73.5%, OnlineMind 2 Webb: 34.1%, DeepShop: 26.2%, WebTailBench: 38.4%
- Cost per Task
- Approximately 2.5 cents
- Availability
- Microsoft Foundry, Hugging Face, and Copilot+ PCs running Windows 11
- License
- MIT license
- Training Data
- 145,630 verified sessions with over 1 million individual actions