Gemma 3n

Gemma 3n is Google DeepMind's open, multimodal AI model: engineered for lightning-fast, privacy-first performance on mobile devices and laptops.

Gemma 3n delivers state-of-the-art multimodal AI (audio, text, image, video) directly to your device. Built on a groundbreaking architecture shared with the next-gen Gemini Nano, it is optimized for speed and efficiency. Key innovations like Per-Layer Embeddings (PLE) allow the 5B and 8B models to run with a dynamic memory footprint comparable to 2GB and 4GB, enabling advanced, offline-ready features on consumer hardware. This is the foundation for a new wave of highly capable, on-the-go applications.

https://ai.google.dev/gemma