If OpenAI resolved the Riemann hypothesis by finding a nontrivial zero at 0.50000003 + 3531696584231.17843174i, that would be very cool and probably very impactful. But it might not necessarily demonstrate model capabilities beyond those which have already been demonstrated, especially if the session that found it was one of a thousand launched to explore different areas of the problem space.

More generally, do you expect that there's some capability threshold where people will no longer study or analyze AI model outputs, and instead just sit there slack jawed saying "so cool!" every time OpenAI announces novel ones? I don't really understand why that would be or why someone would want that. If you're interested in the pure experience of a complex machine outputting satisfying results, I'd recommend getting into sports cars.