Google launches Gemma 4: The most powerful open source AI family yet

On April 2, 2026, Google released Gemma 4, the latest generation of its open AI models. It's the most advanced open source family Google has released, built on the same research and technology as Gemini 3.

Gemma 4 comes in four sizes: Effective 2B, Effective 4B, 26B Mixture of Experts, and 31B Dense. The largest model ranks among the strongest open models in the world right now. All four sizes support image and text inputs, while the two smaller variants also support native audio input, enabling speech recognition and audio/video analysis on edge devices.

The Gemma family has been downloaded over 400 million times since the first generation. With the 31B Dense model, Google delivers performance that competes with closed models, but under the Apache 2.0 license, giving full freedom to use, modify, and commercialize.

For developers and companies, this is directly relevant: Gemma 4 is available today via Google AI Studio, Hugging Face, and Google Cloud. Smaller variants run on Raspberry Pi and NVIDIA Jetson Orin Nano, making on-device AI more accessible than ever.

For CIOs evaluating open source alternatives to closed models, Gemma 4 is a concrete and compelling option that gives full control over data and infrastructure.