For api / switching based on speed: do you mean specifically between the identical or nearly models like gpt-4 and gpt-4-0613, or do you mean across non identical models too?