Gemma Open Models

New release

Gemma 2

Gemma 2 adds two new powerful and efficient additions available in 9 billion and 27 billion parameter sizes, with safety advancements built in.

Gemma 1

Gemma models are lightweight, text-to-text, decoder-only large language models, trained on a massive dataset of text, code, and mathematical content for a variety of natural language processing tasks.

RecurrentGemma

RecurrentGemma is a technically distinct model that leverages recurrent neural networks and local attention to improve memory efficiency.

PaliGemma

PaliGemma is an open vision-language model inspired by PaLI-3, leveraging SigLIP and Gemma, designed as a versatile model for transfer to a wide range of vision-language tasks.

CodeGemma

Harnessing the foundation of our original pre-trained Gemma models, CodeGemma brings powerful code completion and generation capabilities in sizes fit for your local computer.