Gemma Downloads Pass 1 Billion, 100,000+ Variants

Gemma downloads pass 1 billion, and Gemmaverse variants top 100,000

Google announced on August 20, 2026 that the Gemma family of open models has passed 1 billion downloads. The announcement is a blog post titled "Inside the Gemmaverse: Celebrating one billion Gemma downloads", written by Clement Farabet and Olivier Lacombe of Google DeepMind. Google says the aim at launch was to let developers build responsible AI applications anywhere, and that Gemma models soon showed they could be deployed on local devices, on edge infrastructure and even in space.

The second figure is about developers. In the past two years they have published more than 100,000 Gemma model variants, an ecosystem Google calls the Gemmaverse. Google says what people build with Gemma matters far more than the download count.

The figures in Google's announcement

Figure	Value
Gemma family downloads	Over 1 billion
Gemma variants published by developers in two years	Over 100,000
Gemma Challenge projects submitted on Kaggle	Over 1,600
Aarogya Setu 2.0 downloads on Android	Over 100 million

The Gemma Challenge on Kaggle is a recent contest for projects aimed at real-world problems. In the August 20 post, Google says the community submitted over 1,600 projects and will announce the winners very soon. For another open model family's download figure, see Qwen Downloads: Checking the 3 Billion Figure; GLM 5.2 benchmarks, pricing, and Claude Code setup covers another open-weight release.

Awesome Gemma on GitHub: the official Gemmaverse directory

Awesome Gemma is a curated GitHub repository that Google launched with the milestone post. Google says it will serve as the official directory for the Gemmaverse and will feature the best community projects, fine-tunes, tutorials and developer tools.

Google gives its reason in one sentence: with over 1 billion downloads and so many projects being built every day, it wants to help the community share, discover and collaborate. If you are looking for community fine-tunes, tutorials or tooling for Gemma, this is the listing Google itself calls official.

Gemma open model in orbit: NASA, Satlyt and Starcloud satellites

NASA, Satlyt and Starcloud are running Gemma directly in orbit, Google says. Google lists three jobs for the model on board:

• Onboard image analysis.
• Optimizing scarce downlink bandwidth, meaning the limited capacity for sending data from a satellite down to Earth.
• Routing intersatellite communications, meaning the links between satellites.

Google says the three organizations together prove that Gemma can deliver complex reasoning in some of the most constrained and extreme environments imaginable. That is the company's own reading of its partners' work, not an independent test.

C2S-Scale on Gemma: a cancer pathway verified in living cells

C2S-Scale is an AI model that researchers from Yale and Google built on Gemma. It is designed to interpret the "language" of single cells.

According to Google, C2S-Scale discovered a novel cancer therapy pathway that was then verified in living cells. Google adds that this is the first time an AI system has produced novel mechanistic therapeutic pathways that were verified in living cells.

MedGemma in clinics and Gemma 4 in Aarogya Setu 2.0

MedGemma is Google's domain-specific medical model, and the company says organizations around the world use it to build real-world clinical applications. The post names two examples: supporting outpatient triage at the All India Institute of Medical Sciences (AIIMS), and systems that support frontline health workers in rural Uganda.

India's National Health Authority integrated Gemma 4 and the Google open-source Medical Data Toolkit into Aarogya Setu 2.0, an app with over 100 million downloads on Android. Gemma 4 turns complex medical reports into standardized digital formats, which helps citizens manage their health data and share it securely across providers. Google says this shows how developers can deploy its open models to handle critical information at massive scale.

DolphinGemma and dolphin vocalizations: an open problem for Gemma

DolphinGemma is a specialized AI model that researchers developed in collaboration with Georgia Tech and the Wild Dolphin Project. It processes complex dolphin vocalizations to predict sound sequences, that is, the order in which one sound follows another.

Google describes the work as ongoing. It says researchers are actively iterating on this complex challenge and that it is just the beginning.