> I just want some sort of mechanism where the model can "think" longer about harder problems.

There is such a mechanism - multiple rounds of prompting. You can implement diverse patterns (chains, networks) of prompts.