I don't notice any improvement with GPT-5. If they truly made advancements, they shouldn't have removed the legacy models. I think they introduced this model to cut server costs by limiting the use of O3 and other heavy models. I'm considering switching to other platforms since, even after paying, they pull these shitty moves. Only the Pro ($200) model gets access to the legacy models.
The graphs they show in their papers are absolutely funny af.
They’re heavily marketing it as an absolute improvement, but I don’t think we, (public), will get much better models even if they create them, due to the cost. It’s like a cold war between countries now, so they’ll probably keep the better models for themselves if they manage to develop them.