> Set your model and effort level before you start. Changing either one mid-conversation can bust your prompt cache, which can increase token cost.

I know we supposed to do this but is there any particular reason why such things cannot be supported? I thought its running on same model just different settings like reasoning. This would be super useful.