At Google I/O 2025, the tech giant introduced Gemma 3n, the newest addition to its family of “open” AI models — and this one is designed to run directly on mobile devices. Starting Tuesday, developers can preview the lightweight yet powerful model, which supports audio, text, images, and video processing.
According to Google, Gemma 3n is optimized to work “smoothly” on phones, laptops, and tablets—even those with less than 2GB of RAM. This means users can now run advanced AI models offline without relying on cloud computing, making them cheaper and more private.
“Gemma 3n shares the same architecture as Gemini Nano, and is engineered for incredible performance,” said Gus Martins, Product Manager for Gemma, during the keynote presentation.
Running models locally on-device has become an exciting area in AI development, offering lower latency and enhanced privacy. Gemma 3n is part of Google’s push to make high-performance AI more accessible across a broader range of consumer devices.
MedGemma: AI for Health Apps
Google also unveiled MedGemma, its most capable open model for understanding health-related images and text. Released under its Health AI Developer Foundations program, MedGemma is meant to help developers create smarter, more adaptive healthcare apps.
“MedGemma works great across a range of image and text applications, so that developers can adapt the models for their own health apps,” said Martins.
SignGemma: AI for Sign Language Translation
Looking ahead, Google previewed SignGemma, a model designed to translate sign language into spoken-language text — with a focus on American Sign Language (ASL) and English.
“SignGemma is the most capable sign language understanding model ever, and we can’t wait for you — developers and deaf and hard-of-hearing communities — to take this foundation and build with it,” Martins added.
SignGemma is aimed at enabling a new wave of accessibility-focused applications for the deaf and hard-of-hearing community.
A Note on Licensing
While the Gemma models have been downloaded tens of millions of times, some developers have criticized Google’s non-standard licensing terms, which they say may complicate commercial use. Still, the models continue to gain traction due to their flexibility and performance.
Gemma 3n and its companion models represent Google’s latest attempt to democratize AI development while keeping performance high and hardware requirements low.
