The worse piece of news is that the performance of thinking and non-thinking models is inverted depending on whether the problem is simple or medium, which impl

The worse piece of news is that the performance of thinking and non-thinking models is inverted depending on whether the problem is simple or medium, which implies that you need to know the problem's complexity to choose the best model. Ouch.