Luckily for programmatic or logic following, smaller models tend to do better, it can be surprising at first to see the more expensive models do worse at a task but it’s true.