How Google’s 2.3B Gemma 4 Model Rivals 70B Giants on Just 1.5GB of RAM

Google has released Gemma 4, a 2.3B-parameter open-source AI model.

It reportedly matches some 70B-model tasks while using less than 1.5GB of RAM.

The model supports text, vision, audio, 140+ languages, and a 128K context window for offline multimodal use.

Its low-memory design could bring capable local AI to cheap smartphones and edge devices, improving privacy and deployment flexibility.