Google said its Gemma family of open-weight AI models has surpassed 1 billion cumulative downloads, with outside developers creating more than 100,000 fine-tuned or modified variants since the models first launched.
Unlike Google's flagship Gemini models, which run primarily through the company's own cloud and apps, Gemma models are released with weights that developers can download, modify and run on their own hardware. That approach has made Gemma popular with researchers, hobbyists and startups that want more control over cost and customization than a hosted API allows.
The milestone underscores how central open-weight models have become to the broader AI ecosystem, even as the largest labs — including Google itself — continue to keep their most capable systems closed. Meta's Llama family has pursued a similar open strategy, and the two companies now effectively compete for developer mindshare in addition to enterprise cloud customers.
Google has continued to release specialized Gemma variants tuned for tasks like coding and on-device use, betting that a thriving developer ecosystem around the models will feed back into demand for its cloud and hardware.
