Google's Gemma models passed 150M downloads and 70K+ variants on Hugging Face, after its February 2024 launch; Meta's Llama passed 1.2B downloads in April 2025
I may have little love for this ‘revolution,’ but there are bright spots. techcrunch.com/2025/05/12/g... X: Omar Sanseviero / @osanseviero : Gemma just passed 150 million downloads and over 70k variants on Hugging Face🚀🚀🚀 What would you like to see in the next Gemma versions?
Context & Ripple Effects
Gemma’s adoption milestone follows Google’s March release of Gemma 3 models designed to run on a single GPU, extending the project’s emphasis on making capable models usable with relatively modest hardware. The 70,000-plus Hugging Face variants indicate that distribution is producing a sizable downstream developer ecosystem, not just base-model downloads.
The comparison with Llama’s larger download total puts Gemma’s progress in a competitive open-model market where reach matters, but commercial-use constraints in “open” model licenses remain a practical consideration for developers choosing a foundation model.
First-order effects
- Google gains a visible adoption signal for Gemma: 150 million downloads and a large variant base strengthen its case for continued model releases and developer support.
- Developers and model hosts have a broader Gemma-derived ecosystem to evaluate, while Meta retains a substantial scale lead on the stated download measure with Llama’s 1.2 billion total.
Second-order effects
- The size of the variant ecosystem raises the value of compatibility, tooling, and documentation around Gemma, making it harder for rival open-model families to win developers solely through a new base release.
- Download counts become a more prominent competitive metric for Google and Meta, though licensing terms can still limit how readily organizations turn community experimentation into commercial deployment.
Third-order effects
- If variant creation keeps concentrating around a small number of model families, open-model competition may increasingly hinge on ecosystem distribution and deployability rather than benchmark performance alone.
- The gap between “open” availability and permitted commercial use could become a lasting point of differentiation, giving buyers more leverage to demand clearer terms as they compare model ecosystems.
The trend: Open-model competition is shifting toward distribution, derivative-model ecosystems, and usable licensing alongside raw model capability.