Yep, if fishing for compliments from a robot doesn't feel like the future, I don't know what does.

Certainly very impressive, and somewhat scary. $200 a month is a lot cheaper than my salary! Can it synthesise outputs too, give it a novel problem (one clearly not present in its training data) and get a reasonable output?
This qualifies as a novel problem, I think, given that the idea was to act as a brainteaser, and given that it can't possibly be in the training corpus. If you read the solution threads, they are full of responses from people who are (by appearances anyway) working EEs or at least EE students. Not all of them get the answer right, and some of them get pretty argumentative.
One excellent benchmark is the
geometry quiz from last May. Most respondents got it wrong, and some got downright pissy about it. Previous OpenAI models didn't come close to getting it, and o1-pro didn't get it either, but it used the same flawed reasoning as some of the humans to get the same wrong answers. If I argue with it enough, I can convince it of the correct answer. I wouldn't be surprised if the next major release has no trouble with it.
IMHO even the $200/month model is still a work in progress, and we are still a long way from being able to trust these models implicitly. But you can treat it like an intern or junior employee who is
very good at design planning, sanity checks and general brainstorming.
I used to double-check ChatGPT results against Google Gemini, but there's no point doing that with o1-pro. It has no peers right now.