Linking to he announcement (which links to his PDF) would probably be more useful.

Introducing Gemma 3: The most capable model you can run on a single GPU or TPU

https://blog.google/technology/developers/gemma-3/