A manufacturer wants a Google-built model whose weights its engineers can download, adapt to their own data, and run on servers in their own facilities. Which Google model family fits this requirement?
Choose one.
Gemma is the clearest expression of Google's open approach: open models whose weights organizations can download, adapt and run where they choose.
The requirement to download weights and adapt them rules out models whose weights Google keeps. Gemini, Imagen and Veo are Google first-party models offered as managed models (Gemini can even be deployed on Google Distributed Cloud in a customer's facility, still without its weights being released), while Gemma is released as open models built from the same research. That combination of Google-built quality and deployment freedom is what the open approach offers a business.
- Spot the deciding requirement: downloadable weights the engineers can adapt and run themselves.
- Separate managed first-party models (Gemini, Imagen, Veo) from open-weights models.
- Pick the Google model family released with open weights.
Exam tip: Need Google-built weights you can run yourself? That is Gemma, the open part of Google's model lineup.
Google Cloud's Gen AI Strengths: Platform, Infrastructure, and Openness — the lesson that teaches this.