Google Unveils Gemma 3n: Powerful AI That Runs on Phones, Laptops, and Tablets

Google is continuing to expand its line of “open” AI models with new announcements made at Google I/O 2025. Gemma 3n, a lightweight version designed to function well on computers, tablets, and smartphones, was introduced by the firm. Now available in preview, Gemma 3n supports a wide range of data types, including text, images, video, and audio.

These types of compact models, which operate directly on devices without needing cloud-based processing, have become increasingly popular. They’re more cost-effective than large-scale models and offer stronger privacy protections by keeping user data local instead of transmitting it to remote servers.

Gus Martins, Product Manager for Gemma at Google, said during the I/O keynote that Gemma 3n supports devices with less than 2GB of RAM. “It uses the same architecture as Gemini Nano,” Martins explained, “and is optimized to deliver excellent performance even on limited hardware.”

Alongside Gemma 3n, Google also revealed MedGemma, launched as part of its Health AI Developer Foundations initiative. According to the company, this model is their most advanced open system for processing health-related images and text. “MedGemma is a versatile collection of models designed for multimodal medical applications,” Martins stated, adding that it can be adapted to a variety of healthcare tools.

Google also gave a glimpse of SignGemma, an upcoming model built to convert sign language into written text. Initially focused on translating American Sign Language into English, SignGemma is intended to support new technologies for the deaf and hard-of-hearing community. “It’s the most advanced sign language model we’ve ever developed,” said Martins, encouraging developers to explore its potential.

Although Gemma models have been criticized for their unique licensing structure – which some claim complicates commercial use – this hasn’t slowed their adoption. Collectively, these models have been downloaded tens of millions of times, showing strong demand across the developer community.