The token efficiency improvement might be underrated. If the model solves tasks with fewer tokens, that directly translates into lower cost and faster responses for anyone building on the API.