Yeah this is what I was thinking, or maybe use smaller models to work on the prompt then fire it off to the biggie while you do something else.