Google Introduces Gemma 3n: An AI Model Built for Low-Power Devices

At Google I/O 2025, Google unveiled Gemma 3n, the latest addition to its Gemma line of open AI models, designed to run efficiently on low-resource devices like smartphones, laptops, and tablets. The model supports multiple modalities—text, audio, images, and video—and can function offline with less than 2GB of RAM, promoting better data privacy and lower usage costs. According to Google Product Manager Gus Martins, Gemma 3n shares its architecture with Gemini Nano, ensuring high performance despite limited device resources.

Alongside Gemma 3n, Google also announced MedGemma, its most advanced open model for interpreting health-related texts and images. Released under the Health AI Developer Foundations program, MedGemma aims to help developers build healthcare applications that leverage multimodal understanding. Additionally, the upcoming SignGemma model is designed to translate American Sign Language into spoken-language text, offering new opportunities for accessibility-focused apps for deaf and hard-of-hearing users.

Despite criticism surrounding Gemma’s custom licensing, which some say complicates commercial deployment, the models have been downloaded tens of millions of times. Overall, the announcements demonstrate Google’s commitment to democratizing AI across sectors like health and accessibility, while making it available for low-spec devices through edge processing.