Google announced on 20 August that its open Gemma model family has passed one billion downloads, alongside over 100,000 community-published Gemma variants and more than 1,600 submissions to the recent Gemma Challenge on Kaggle. The post also launches an "Awesome Gemma" repository on GitHub as a curated directory for what Google calls the Gemmaverse.
What the post actually claims
Google is explicit that the count is not the point: "What matters far more than the download count is what the community is building with them." The examples given are specific. Teams at NASA have put Gemma in orbit for onboard image analysis, bandwidth optimisation and intersatellite routing. India's National Health Authority integrated Gemma 4 and Google's open-source Medical Data Toolkit into Aarogya Setu 2.0 to turn medical reports into standardised digital formats.
What the common framing gets wrong
The billion and the hundred thousand are not independent numbers. Every community variant is itself a downloadable artefact on the same hubs, and its pulls count toward the same total. An ecosystem that produces 100,000 re-uploads is an ecosystem that mechanically inflates a download counter — which may be why Google put the ecosystem claim ahead of the number. Separately, the widely circulated figure that Gemma 4 alone accounts for 300 million of the downloads does not appear anywhere in Google's post; it is being back-filled by aggregators. And the section headed "Empowering 100 million citizens in India" rests on Aarogya Setu 2.0 having over 100 million Android installs — that is the app's install base, not a count of people using Gemma.
Why the counter still matters
Download totals are the only public traction metric open-weights families have. There is no revenue line, no seat count, no API meter. That makes the number both the industry's standard yardstick and its softest one, and it is why the comparison everyone wants — Gemma against Llama or Qwen — cannot be made cleanly from published figures.