Google’s Gemma AI Model Can Now Run on Phones

Google is expanding its Gemma family of open AI models, introducing Gemma 3n, designed for optimal performance on phones, laptops, and tablets. The announcement was made at Google I/O 2025, where the company highlighted the model’s ability to handle audio, text, images, and videos. Gemma 3n is available in preview starting today.

The trend toward efficient, offline-capable AI models has been growing, driven by the need to reduce costs and enhance privacy. Unlike large models that rely on cloud computing, these models process data locally, eliminating the need to transfer data to remote servers.

Gus Martins, Product Manager at Google, mentioned during his keynote that Gemma 3n can operate on devices with less than 2GB of RAM. He emphasized that Gemma 3n shares the same architecture as Gemini Nano and is engineered for incredible performance.

In addition to Gemma 3n, Google is launching MedGemma through its Health AI Developer Foundations program. MedGemma is designed for analyzing health-related text and images. According to Martins, MedGemma works across a range of image and text applications, enabling developers to adapt the models for their own health apps.

Google is also working on SignGemma, an open model designed to translate sign language into spoken-language text. The company aims to empower developers to create new apps and integrations for deaf and hard-of-hearing users. SignGemma, according to Martins, excels at American Sign Language and English translation, making it the most capable sign language understanding model ever.

It’s worth noting that Gemma’s custom, non-standard licensing terms have faced criticism. Some developers consider the commercial use of these models risky due to these restrictions. Despite this, Gemma models have been downloaded tens of millions of times.

Updated 2:40 p.m. Pacific: Added several quotes from Gemma Product Manager Gus Martins.