I don't get that from the spec. Closest thing I could find in the Writing and Assistance APIs spec [1] (which the Prompt API refers to) says:
> [It] allows the user agent to prompt the user for permission
But that's talking more about initiating the 'download' process more than choosing the model.
I think Chrome wants the browser to be 'opinionated' about the model used however, rather than letting the webpage pick, at least for now.
[1] https://webmachinelearning.github.io/writing-assistance-apis...