Say hello to the expanding family of Google's open AI models, Gemma! At the Google I/O 2025 event, Google introduced its latest addition, Gemma 3n. This new model is customized to work efficiently on smartphones, laptops, and tablets. Starting from Tuesday, you can get a preview of Gemma 3n, a multi-talented AI that can work with text, images, audio, and video content, according to the folks at Google.

In the world of AI, there's been a rising interest in models that are efficient enough to function offline without the need for cloud computing. They're not only more cost-effective than larger models but also advocate privacy by eliminating the necessity to send data to far-flung data centers.

The Gemma 3n is made to run on devices with less than 2GB of RAM, according to Gemma Product Manager Gus Martins during the event. In his words, "Gemma 3n shares the same architecture as Gemini Nano, and is engineered for impressive performance."

As part of its Health AI Developer Foundations initiative, Google intends to launch MedGemma despite revealing Gemma 3n. MedGemma stands as Google's most proficient open model for scrutinizing health-related texts and images.

"MedGemma is our gift to developers who can tailor these models for their health apps. It offers fantastic results in a wide variety of image and text applications," said Martins.

The future also sees the arrival of SignGemma - a pioneer open model aimed at translating sign language into written text. With SignGemma, Google aims to empower developers to invent new apps and upgrades catering to the needs of the hearing-impaired community.

"SignGemma leads the pack as a model trained for sign language translation, especially proficient in American Sign Language and English," commented Martins, "It's the most capable sign language understanding model we've ever created, and we're excited for developers and the deaf community to explore and build upon this platform."

However, a word of caution: there's been some controversy surrounding Gemma's non-standard licensing terms, which some developers claim make commercial use of the models a precarious venture. Despite the uncertainty, Gemma models have been downloaded by developers millions of times collectively.