Google DeepMind releases Gemma 4 under Apache 2.0
Google DeepMind announced Gemma 4, a new family of open models built for reasoning and agentic workflows, released under Apache 2.0.
Google DeepMind has announced Gemma 4, a new family of open models designed to run on user-owned hardware. According to the company, the models are built for advanced reasoning and agentic workflows, and are being released under an Apache 2.0 license.
The Apache 2.0 terms are the notable piece here for builders and buyers. Unlike the custom community licenses attached to many recent open-weight releases, Apache 2.0 is a standard permissive license that allows commercial use, modification, and redistribution without bespoke usage restrictions. For teams evaluating which open models to deploy behind their own agents or reasoning pipelines, licensing friction has become a gating factor alongside raw capability. Gemma 4 positions Google DeepMind against Meta's Llama family and Mistral's open releases in the segment of the market where enterprises want to self-host rather than route workloads through a hosted API. Further technical details were promised in the accompanying thread from Google DeepMind.