Google has released Gemma 3n, a compact AI model that can be run on smartphones

At the I/O 2025 conference, Google announced a new artificial intelligence model, Gemma 3n, which can run directly on mobile devices, laptops, and tablets, TechCrunch reports.

The company says the model is capable of processing text, images, audio, and video, and can even run on devices with less than 2GB of RAM. It is built on the same architecture as the Gemini Nano and is designed to work locally without connecting to the cloud. This reduces computing costs and at the same time increases privacy, as user data is not transmitted to external servers.

Google also unveiled MedGemma, an open model for analyzing medical images and text. It has become part of the Health AI Developer Foundations program and will help developers build medical applications with deeper content understanding.

Another major announcement was the SignGemma model, which translates sign language into text. It currently supports American Sign Language (ASL) and English best. Google hopes that SignGemma will help developers create useful tools for people who are hearing impaired.

Gemma models have a complicated licensing system, and Google's terms are different from typical ones, which makes some developers consider their use in commercial projects potentially risky. Despite this, Gemma has already been downloaded tens of millions of times.

Gemma product manager at Google, Gus Martins, emphasized that the company will continue to develop these models and ensure their openness to the developer community.

By the way, in addition to Gemma 3n, at the I/O conference, Google also introduced a new AI Ultra tariff plan for $250 per month, a real-time translation function in Google Meet, a new AI tool for creating interfaces Stitch, smart glasses in collaboration with Xreal, automatic creation of strong passwords in Chrome, and a new generation of image generator Imagen 4.