Gemma 3

Run Gemma 3 inference locally on Linux through a snap that supplies an environment optimized for the model. Ubuntu publishes it as a generative-AI package that turns the host machine’s available hardware into a local inference service. The snap detects the system’s CPU, GPU, or NPU and chooses matching runtime and model-weight optimizations automatically.

Users interact with Gemma 3 through its command-line interface, including the gemma3 --help command for usage guidance. Applications can also connect through a standardized local API, giving developers a consistent way to add AI features without tuning separately for each hardware type. Typical uses include supplying local model capabilities to IDEs, chatbots, and media-editing software.

The host must run Linux with snapd, and the necessary hardware drivers must be installed before inference begins. The package supports amd64 and arm64 architectures, making it suitable for compatible desktop, workstation, server, and other Linux systems. Gemma 3 is freeware, while use of the model is governed by Gemma’s Terms of Use; licenses for bundled software are available inside the snap at /snap/gemma3/current/usr/share/doc. The current version is 2.0.0-beta.14+a169df7.