I have found that higher reasoning doesn't always produce better results. Models often tend to overthink and get themselves in a bind. And tasks just take longer for no reason.