So someone used a small model with many requests & a massive amount of tokens, to eventually beat a large model with a single request & tiny amount of tokens. .....Cool? Were they not expecting that to work?