Regardless of which entry point is used to access Google's AI services, all inference requests ultimately converge on Google Cloud's unified inference cluster, running the same version of model weights and sharing the same safety filter pipeline and sampling configuration.