Google DeepMind says its Gemma family of open models has surpassed 1B downloads and developers have published 100K+ Gemma model variants over the past two years
From outer space to underwater, here's how developers are bringing a positive impact with our open models.
Context & Ripple Effects
Gemma began as Google's smaller open-source 2B and 7B models, then expanded into specialized releases such as CodeGemma and RecurrentGemma for coding and faster inference. By May 2025, the family had reached 150 million downloads and more than 70,000 variants; the new figures show that developer adaptation has continued to outpace the base-model catalog.
Google also moved Gemma 4 to an Apache 2.0 license for reasoning and agentic workloads, a meaningful change after developers had raised concerns about restrictions in some ostensibly open-model licenses. The scale of variants makes the licensing and distribution terms increasingly consequential to the ecosystem around Gemma.
First-order effects
- Google DeepMind gains a large installed base around Gemma, while developers have a broader pool of more than 100,000 community-built variants to reuse and adapt.
- Gemma 4's Apache 2.0 licensing gives developers a clearer basis for building on the latest family after earlier debate over commercial-use constraints.
Second-order effects
- Meta's Llama and other open-model providers must compete not only on their base models but on the size and usability of their derivative-model communities; prior coverage put Llama at 1.2 billion downloads in April 2025.
- Organizations selecting open models gain more implementation choice, but must evaluate variants and the underlying license rather than treating “open” as a uniform commercial-use guarantee.
Third-order effects
- Open-model competition is increasingly organized around ecosystems of fine-tuned and specialized derivatives, where the base-model provider's licensing and distribution choices shape downstream adoption.
- If this pattern persists, model families with permissive, stable terms and active variant communities will hold an advantage over providers whose access conditions create uncertainty for commercial builders.
The trend: Open AI models are becoming ecosystem platforms, with derivative-model volume and license clarity joining raw model capability as competitive assets.