“The O1 model is very good at reasoning and solving LeetCode-style problems but is not as good as Sonnet at understanding a user's rough intent.”