Gemma 3n E4B

Gemma 3n E4B is a 4-billion parameter multimodal small language model (SLM) optimized for high-speed native reasoning across text and vision.

Google's Gemma 3n E4B leverages a hybrid architecture to deliver 4B-parameter efficiency with the performance of larger dense models. Built on the same research as Gemini, this specific 'n' variant focuses on native multimodal processing (processing images and text in a single stream) while maintaining a small enough footprint for local deployment on consumer hardware. It excels in low-latency environments, providing developers with a versatile tool for edge-based visual analysis and complex instruction following without the overhead of 7B+ parameter alternatives.

https://blog.google/technology/google-deepmind/google-gemma-3-ai-model/