its really good at planning however it does have a very short context window due to large portion of the context window used during thinking process. That is probably what's holding the output quality back. To tackle this i generally start with o1 preview for the base approach and a draft, use o1 to add/remove stuff you don't need don't exceed 1 chat, keep on editing it, annoying but ik we don't rlly have a huge context window). Continue with 4o/claude from that point onwards