Yes, this is important. Gpt 5 and o3 were ~ equivalent for a one shot one file task. But 5 and codex-5 can just work for an hour in a way no model was able to before (the newer claudes can too)
I use the newer claudes and letting them work for 1 hour leads to horrible code over 50% of the time that does not work. Maybe I am not the target person for agentic tasks, all I use agents for is to do product searches for me on the internet when I have specific constraints and I don't want to waste an hour looking for something.
