Gemma is Google's family of lightweight, open-weight AI models, built from the same research and technology behind Gemini but released for anyone to download, run locally, and fine-tune under the Apache 2.0 license. Unlike Gemini, which you access as a hosted product, Gemma is meant to run on your own hardware — from a phone to a high-end workstation — or through your own cloud deployment.

Gemma 4, released April 2, 2026, ships in four sizes: E2B (2B parameters, tuned for phones), E4B (4B, for edge devices), a 26B mixture-of-experts model with 3.8B active parameters for consumer GPUs, and a 31B dense model for workstations. All sizes handle multimodal text-and-image input, the edge models add audio, and every size supports long context windows up to 256K tokens along with native support for agentic, tool-using workflows.

Gemma is completely free to download and use under its open license. It runs locally with a single command via Ollama, is available through Hugging Face and Kaggle, works in Google Colab with PyTorch or JAX, and can be deployed at scale through Vertex AI, Cloud Run, GKE, or TPU-accelerated serving for production or regulated workloads.