I feel like the gpt-5-pro model has been dumbed down, and the reasoning time is much shorter each time.