Google's open model family Gemma has passed one billion cumulative downloads. The milestone comes roughly two and a half years after the models first appeared in February 2024, and developers have published more than 100,000 variants of their own. On August 20, Google marked the occasion by detailing where the models are actually being used, and announced a new GitHub repository called Awesome Gemma to gather the community's work in one place.
From 900 Million to 1 Billion in About a Month
Gemma is Google's line of open-weight models. They run on cloud servers, but also on a laptop or an edge device, which means inference can stay entirely inside your own environment without calling an external API.
Cumulative downloads were reported at 900 million as of July 2026, so the last 100 million arrived in a little over a month. Much of that momentum comes from Gemma 4, the newest generation released in April 2026. It ships under the Apache 2.0 license, which keeps the terms straightforward for commercial use.
The composition of that number is more interesting than the number itself. Developers have published over 100,000 variations tuned for particular languages, tasks, and hardware targets. Google calls this sprawl the Gemmaverse.
Running Inference in Orbit
The most unusual deployment is in space. Teams at NASA, satellite startup Satlyt, and orbital computing company Starcloud are running Gemma directly in orbit.
The work covers onboard image analysis, optimization of scarce downlink bandwidth, and routing for intersatellite communications. When the pipe back to Earth is extremely narrow, the ability to decide on the spot what is worth transmitting shapes the entire mission. That complex reasoning holds up in an environment this constrained in both compute and power says a lot about the practical value of shipping weights that anyone can place on their own hardware.
Built Into an App Used by 100 Million People in India
Back on the ground, India offers the largest example. The country's National Health Authority integrated Gemma 4 and Google's open-source medical data toolkit into Aarogya Setu 2.0, an app with more than 100 million downloads on Android alone.
Its job is converting inconsistently formatted medical reports and clinical records into standardized digital formats. That lets citizens manage their own health data and share it securely across providers. Being able to process sensitive personal health information without shipping it to an external API is likely part of why an open model was chosen.
A Cancer Pathway and MedGemma
On the research side, Google pointed to C2S-Scale, developed jointly by researchers at Yale and Google. It interprets the behavior of single cells as a kind of language, and it is built on Gemma. The model identified a novel therapeutic pathway for cancer that was then confirmed in living cells. Google describes this as the first time an AI system produced a novel mechanistic therapeutic pathway that was verified in live cells.
MedGemma, the medicine-focused derivative, is in use across several countries. It supports outpatient triage at the All India Institute of Medical Sciences and underpins systems for frontline health workers in rural Uganda. The range spans workflow management and patient communication through to diagnostic and treatment support.
DolphinGemma and the Sounds Under the Sea
A stranger project is DolphinGemma, a collaboration with Georgia Tech and the Wild Dolphin Project. The model processes complex dolphin vocalizations and predicts sound sequences.
Nobody has solved interspecies communication yet, and the researchers are still iterating. What stands out is that a general-purpose open model can now serve as the foundation for research this far outside its original design brief, which is itself a marker of how much has changed in two years.
Awesome Gemma Pulls the Scattered Work Together
The community side is busy too. The Gemma Challenge on Kaggle drew more than 1,600 submissions aimed at real-world problems, and winners will be announced shortly.
At the same time, 100,000 variants and a steady stream of new projects had become scattered and hard to search. The new Awesome Gemma repository is positioned as the official directory for the Gemmaverse, organizing notable community projects, fine-tuned models, tutorials, and developer tools. It is an attempt to fix the entry-point problem that tends to follow open models around: knowing where to start.
Summary
Gemma has reached one billion cumulative downloads with more than 100,000 published variants. Image analysis in orbit, a health app serving 100 million people in India, cancer pathway discovery, and dolphin vocalization research make the spread of use cases more striking than the headline figure. Whether Awesome Gemma works as a usable map of all that output is the next thing worth watching.
