Thank you, that's very helpful. I think I could get close to that in some coding sessions where I'm running multiple in parallel but I suspect it's very very rare. Even with token efficient gpt5-codex my OpenAI bill is quite high so I think I will switch to Pro now.