I have to agree. I have a set of oddball questions that I ask every new language model, almost all of which have been failed by every model, and o3 mini gives very credible answers to basically all of them.

This is a little less impressive than it sounds, because my questions involve a weird mix of unexpected scenarios and need for nontrivial analysis, and that's what o3 is supposed to do. But, at least in my hands, it really does do it.