Google releases DeepMind's Gemma 4 AI model under Apache 2.0

Technology Oct 06, 2026

Google just rolled out Gemma 4, its latest AI model from DeepMind, and this time it's fully open-source under Apache 2.0.

What's cool? You can run it locally, perfect for anyone working with sensitive information, like in health care.

Unlike earlier versions with more limits, Gemma 4 is built to be way more accessible for all kinds of projects.

Gemma 4 variants for servers, phones

Gemma 4 comes in four flavors: two (26B and 31B) made for powerful servers, and two (E2B and E4B) designed to work smoothly on phones or even tiny devices like Raspberry Pi.

It handles advanced tasks offline with low lag and supports more than 140 languages, so it's ready to go global.

Thanks to partners like Qualcomm, real-time AI on your device is now a real thing.