Google DeepMind releases Gemma 4: open-source AI runs locally

Technology Apr 02, 2026

Google DeepMind just rolled out Gemma 4, a new open-source model family that lets you run AI locally; no cloud is needed. That is especially helpful for privacy-focused fields like healthcare.
The four-model set is built to work everywhere, from powerful servers to your phone or smart devices.

Gemma 4 mobile optimized and multilingual

Gemma 4 comes packed with features: smarter reasoning, better security, and the ability to work offline.
It can generate code and process images, and it was trained on over 140 languages, and thanks to partnerships with Qualcomm and MediaTek, it runs smoothly on mobile too.
With more than 400 million downloads since its first release in 2024, making Gemma fully open-source means even more developers can use it worldwide for all kinds of projects.