> we already know it's possible to make a pretty darn clever reasoning model

There's is a problem though: we know that it is possible, but we don't know how to (at least not yet and as far as I am aware). So we know the answer to "what?" question, but we don't know the answer to "how?" question.