Google DeepMind has introduced Gemma 3n, a new multimodal AI model designed specifically for mobile on-device applications. The model significantly reduces RAM usage by nearly threefold, enabling it to run efficiently on devices with as little as 2GB of RAM. Gemma 3n supports complex AI tasks including processing text, images, audio, and video, making it a versatile tool for mobile and edge computing. It achieves faster response times—approximately 1.5 times quicker on mobile devices—through advanced techniques such as layer-wise embedding and key-value cache sharing. The model, which is around 5 billion parameters in size but with memory usage comparable to a 2 billion parameter model, is expected to be integrated into Android and Chrome platforms. This development marks a shift in AI inference from centralized data centers to decentralized, user-end devices, promoting broader accessibility and real-time AI capabilities on smartphones and laptops. Gemma 3n is currently available in early preview.