Has anyone tested multiple models for code and figured out which one gave the best results? GPT4o is pretty decent, but still takes many takes sometimes for very simple scripts. I wonder if there are better alternatives out there.