Term
Gemma 3n
Google's open, mobile-first AI model, available in preview from May 2025 and fully from 26 June 2025. Multimodal (text, image, audio, video), optimised for on-device use; open weights in the E2B and E4B sizes.
Gemma 3n — explained in more detail
Gemma 3n belongs to Google’s open Gemma model family and is designed specifically to run directly on end devices (on-device). Google first introduced the model in May 2025 as a preview; the full release followed on 26 June 2025. Unlike proprietary cloud models, the weights are openly available and can be run locally.
Gemma 3n is multimodal: it processes text, images, audio and video as input and produces text output. It comes in two sizes, E2B and E4B. The “E” stands for “Effective” (effective parameters): the actual parameter counts are roughly 5 and 8 billion, but thanks to memory optimisations the models only need the memory footprint of about 2 and 4 billion parameters at runtime. Gemma 3n thus brings multimodal capabilities previously reserved for cloud frontier models onto mobile and edge hardware.
At launch, a broad ecosystem supported the model, including Hugging Face, llama.cpp, Ollama, LM Studio, MLX, NVIDIA and vLLM.
Example / practical relevance
The on-device focus makes Gemma 3n attractive wherever data should not leave the device or where there is no stable internet connection: voice and image processing directly on the phone, offline assistants, or privacy-sensitive applications. Because the weights are open, the model can be freely fine-tuned and integrated into custom apps — with no per-request API cost.
For smaller teams, this means prototyping multimodal AI features locally without depending on a paid cloud provider.
Distinction from similar terms
Gemma is Google’s open model family and is therefore distinct from the proprietary, cloud-based Gemini models. Within Gemma, the “3n” suffix denotes the mobile-first, on-device-optimised variant of the third generation; alongside it exist regular Gemma 3 models in various sizes. The E2B/E4B size labels refer to the effective memory footprint, not the actual parameter count.
Discover more
GPT-6 Astra Found Questions My First Security Review Missed
I used GPT-6 Astra and Fable 5.1 as independent reviewers for RLS, API and tenant-isolation checks. The useful part was the structured cross-review.
GlossaryGemma 3
Gemma 3 is Googles family of open models released in March 2025. It comes in four sizes (1B, 4B, 12B, 27B), is multimodal from 4B upward, supports over 140 languages and offers a context window of up to 128,000 tokens.
EncyclopediaLLM Model Families at a Glance
Claude, GPT, Gemini, Llama & co. — who builds what, where each family shines, and how to pick the right model for your own use case.