The New Google Gemma AI Model Runs on Phones

At Google I/O 2025 on Tuesday, Google unveiled Gemma 3n, a model built to operate “seamlessly” on phones, laptops, and tablets. Now available in preview, Gemma 3n can process audio, text, images, and videos, according to Google.

In recent years, models capable of running offline and without relying on cloud computing have become increasingly popular in the AI community. These models are not only more cost-effective than large models but also enhance privacy by eliminating the need to send data to remote data centers.

Gemma 3n Can Run on Devices with Less Than 2GB of RAM, Says Product Manager Gus Martins

During a keynote at I/O, Gemma Product Manager Gus Martins announced that Gemma 3n can operate on devices with under 2GB of RAM. He also mentioned, “Gemma 3n shares the same architecture as Gemini Nano and is designed for outstanding performance.“

Alongside Gemma 3n, Google is launching MedGemma through its Health AI Developer Foundations program. The company describes MedGemma as its most powerful open model for analyzing health-related text and images.

Google’s Open Models for Multimodal Health Text and Image Understanding

“MedGemma is our collection of open models for multimodal health text and image understanding,” Martins explained. “It performs well across various image and text applications, allowing developers to tailor the models for their own health apps.”

Additionally, Google is working on SignGemma, an open model designed to translate sign language into written text. This model will enable developers to create new apps and integrations for users who are deaf or hard of hearing.

“SignGemma is a new set of models designed to translate sign language into spoken-language text, with a particular focus on American Sign Language and English,” Martins stated. “It’s the most advanced sign language understanding model to date, and we’re excited for developers, as well as the deaf and hard-of-hearing communities, to use this foundation to build new solutions.”

It’s also worth mentioning that Gemma has faced criticism for its custom, non-standard licensing terms, which some developers believe make commercial use of the models risky. Despite this, developers have downloaded Gemma models collectively tens of millions of times.

Read the original article on: Techcrunch

Read more: Google hits 150M Subscribers, Thanks to AI