I think you can think of Pro approximately as running multiple instances of a non-pro model, than giving all their outputs to a final judge model which combines them into a better one.