Some kind of routing prompts to different models does make sense. But the usecase of saving money on simple prompt.. I think that has only a slight benefit. Fix my typo doesn't use many tokens anyhow.. also model switching still requires carrying over context so it does have some overhead right.