Google has introduced an update to its line of “open” AI models. At its I/O 2025 developer conference, the company launched Gemma 3n, a compact AI model designed to run directly on mobile devices.
Unlike many AI models that require cloud access, Gemma 3n works offline. It supports audio, text, image, and video processing.
This means users can use advanced AI features without internet access or the need for powerful hardware.
Built for Everyday Devices
Gemma 3n can operate on devices with less than 2GB of RAM. According to Google, it shares the same architecture as Gemini Nano, another lightweight model.
During the keynote, Product Manager Gus Martins emphasized performance. He said Gemma 3n is “engineered for incredible performance.” This makes it ideal for smartphones, tablets, and entry-level laptops.
Running AI locally brings several benefits: faster response times and lower data usage. However, the benefit of better privacy takes the cake.
MedGemma
Google also launched MedGemma, its most advanced open model for medical tasks. It was designed to process both health-related images and text.
The model is part of Google’s Health AI Developer Foundations program. Martins described it as a versatile tool. Developers can fine-tune it for medical apps, diagnostics, or image analysis.
AI for Sign Language
Google is also preparing to release SignGemma, a new AI model that translates sign language into written or spoken English. Martins called it “the most capable sign language understanding model ever.”
Currently, SignGemma only works with American Sign Language (ASL) and American English.
SignGemma was made with the intent of powering apps for the deaf and hard-of-hearing community in America. No further information about future language variations has been mentioned.
Licensing Concerns
Despite praise for the technology, some developers have shared concerns about Gemma’s licensing. The models use custom terms that differ from standard open-source agreements.
Ordinarily, these restrictions can make commercial use uncertain. However, user adoption remains high with tens of millions of downloads.
