New Models AI Minute Newsroom 2026-08-22

Google's open models have been downloaded a billion times, and outsiders have built 100,000 versions of them

Google said on 20 August that its Gemma family of open-weight models has passed one billion downloads in roughly two years, and that developers outside the company have published more than 100,000 fine-tuned variants. The announcement came from Clement Farabet, a vice-president at Google DeepMind, and product director Olivier Lacombe, who pointed at where the weights have ended up. NASA, Satlyt and Starcloud are running Gemma on satellites to analyse images in orbit and decide what is worth sending down. India's National Health Authority folded Gemma 4 and Google's open Medical Data Toolkit into Aarogya Setu 2.0, an app with more than 100 million Android installs. A Yale and Google model called C2S-Scale proposed a cancer therapy pathway that was later confirmed in living cells. DolphinGemma, from Georgia Tech and the Wild Dolphin Project, analyses dolphin vocalisations and predicts what sound comes next. Google also opened a directory called Awesome Gemma on GitHub to index the ecosystem.

Why it mattersA billion downloads is a vanity number. A hundred thousand published variants is not. It means the weights are being altered by people Google will never speak to, for problems Google would never have prioritised: a satellite with a bad downlink, a health app in Hindi, a hydrophone in the Bahamas. That is the actual argument for open weights, and Google is making it at a moment when the open-model conversation has largely moved to Chinese labs shipping far bigger models. Gemma's answer is that small and modifiable beats large and impressive for most of the work that actually gets done.

#Health AI #Science & Research

✓ Verified · 3 sources

Google's open models have been downloaded a billion times, and outsiders have built 100,000 versions of them

Related stories