Google DeepMind has open sourced Gemma 4 under Apache 2.0, turning its locally deployable AI model family into a truly redistributable edge and enterprise stack with strong privacy, offline, and multimodal capabilities.
Google DeepMind has made Gemma 4 fully open-source under the Apache 2.0 licence, marking a major strategic shift that transforms the model family into a genuinely redistributable AI stack for enterprise, mobile, and edge deployments. The licensing move is the real breakthrough, replacing the earlier Gemma Terms of Use that limited redistribution and restricted usage categories.
The new framework now enables broad use, modification, commercial deployment, and downstream product integration, sharply distinguishing Gemma 4 from the closed, subscription-based Gemini ecosystem. More importantly, the shift gives developers, OEMs, and enterprises the freedom to build local, on-device AI systems without cloud dependency.
This creates immediate value for privacy-sensitive sectors such as healthcare, where patient data can remain fully on-premise while still benefiting from advanced AI inference and workflow automation. The local deployment model also strengthens data sovereignty compliance across regulated industries.
Gemma 4 is built as a four-model open-source deployment ladder, spanning server-class 26B and 31B models and edge-optimised E2B and E4B variants designed for smartphones, IoT devices, Raspberry Pi-class hardware, and intermittent-connectivity environments.
The release goes beyond a lightweight open model, bringing advanced reasoning, agentic workflows, offline code generation, multimodal image and video processing, variable-resolution support, and speech recognition capabilities in its edge-focused variants.
The Apache 2.0 transition positions Gemma 4 as one of Google’s strongest open-source plays for scalable local AI across datacentre-to-endpoint environments.
