Google Releases Gemma 4 Open AI Models

Google has released Gemma 4, a new family of open artificial intelligence models designed to deliver advanced capabilities on devices ranging from smartphones to high-end workstations. The four models, built on the same research and technology as the company's proprietary Gemini 3, are available under the permissive Apache 2.0 license, marking a shift from the more restrictive terms used in prior Gemma versions. This release enables developers full control over data, infrastructure, and modifications, fostering broader adoption in commercial and research applications.

The Gemma 4 lineup includes two edge-optimized models, Effective 2B and Effective 4B, tailored for low-power hardware such as Android smartphones, Raspberry Pi, and Jetson Orin Nano devices. These smaller variants activate two billion and four billion parameters, respectively, during inference to minimize memory and battery use while supporting near-zero latency operations. They feature native processing of images, video, and audio for tasks like speech recognition, optical character recognition, and chart analysis, with a context window of 128,000 tokens to handle extended inputs such as documents or code snippets.