Google's open Gemma models pass 1 billion downloads and 100,000 variants

Google DeepMind said on August 20, 2026 that its Gemma family of open AI models has passed 1 billion downloads, with more than 100,000 distinct community-built variants published since the models launched in early 2024. Gemma is Google’s line of openly available models that developers can download, fine-tune and run on their own hardware.

The milestone is a marker in the contest for open-model mindshare, where adoption compounds through the derivatives outside developers build. The 100,000-plus variants — fine-tunes and adaptations for specific languages, tasks and hardware targets — are the sign that matters more than raw downloads, because they show developers reshaping the base models for their own use.

Google leaned on deployment examples to make the case. In the announcement, DeepMind vice president Clement Farabet and product director Olivier Lacombe cited uses including NASA and the space firms Satlyt and Starcloud, India’s National Health Authority, research at Yale University, and the Wild Dolphin Project with Georgia Tech.

Those are illustrative, not audited. Cumulative download counts include repeated and automated pulls, and neither the total nor the variant count is independently verified — both are Google’s own figures. They also say little about how many of the deployments run in production versus experiments.

Still, the direction is clear: open models have become a mainstream distribution channel, and the size of Gemma’s derivative ecosystem is the number to watch as Google, Meta and others compete for developers who want to run models on their own terms.