Google's Gemma 4 model goes fully open-source and unlocks powerful local AI - even on phones

Google DeepMind has released Gemma 4 as a fully open-source large language model under the Apache 2.0 license; the article explains licensing changes from prior Gemma releases, describes four model variants (26B, 31B, E2B, E4B) optimized for servers and edge devices, and summarizes capabilities such as advanced reasoning, multimodal input, long context windows, and offline deployment for privacy-sensitive or edge use cases.