Firefox suggests tab groups with local AI (2025) (blog.mozilla.org)

Mozilla has launched a new AI-powered tab grouping feature in Firefox, enhancing users' ability to manage multiple tabs effectively. This innovation, rolled out in early 2025, is significant as it directly addresses user demands for improved organization amid growing concerns about tab overload. Utilizing a local AI runtime, the feature offers two key functionalities: suggesting titles for newly created tab groups and recommending relevant tabs from the current window to be grouped together—all without sending user data to Mozilla. The technical implementation employs a hybrid methodology that combines TF-IDF-based textual analysis and a T5-based generative model, allowing the system to create concise and relevant titles based on distinctive keywords from open tabs. Notably, the model has been optimized for local processing through knowledge distillation and quantization, reducing its size from 1GB to just 57MB while maintaining accuracy. For tab suggestions, a MiniLM embedding model calculates semantic similarities between tabs, with a logistic regression approach improving precision by 18% compared to prior clustering methods. This project exemplifies Mozilla's commitment to enhancing user experience with privacy-centric, efficient AI solutions.