Gemma is Google's family of lightweight open models, built from the same research as Gemini, that anyone can download and run.
In plain words
Gemma gives developers downloadable weights, so they can run, fine-tune and ship the models in their own products, including commercial ones. The first Gemma, in February 2024, came in 2B and 7B sizes. The name comes from the Latin word for precious stone. Besides the main models there are specialised versions, such as EmbeddingGemma 2 for search and ShieldGemma for safety checks.
A quick example
Imagine a school that wants a private homework helper. It downloads a small Gemma model, fine-tunes it on its own course notes, and runs it on its own computers so no student data leaves the building.
Where you'll meet it
- Local and on-device apps.
- Fine-tuned models for specific jobs.
- Community-made variants built on top of Gemma.
Related
Sources
- DocsGemma open models, Google · read 29 Sept 2026
- DocsGemma models overview, Google AI for Developers · read 7 Oct 2026
- DocsGemma 4 model overview, Google AI for Developers · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 7 Oct 2026. Spotted an error? Report it.