Google releases Gemma 4 open models in four sizes under Apache license

Google released Gemma 4, a family of open-source models described as the best open models in the world for their respective sizes. The lineup includes four variants: a 31 billion parameter dense model for raw performance, a 26 billion mixture-of-experts model for low latency and efficiency, and 2B and 4B parameter models designed for edge device deployment. All four are available immediately under the Apache license through Google AI Studio, Hugging Face, and other platforms, with Google encouraging developers to fine-tune them for specific tasks.