> Per the comments on this post, paid versions of at least ChatGPT and Claude produce much better results than the ones I experienced.

I’m surprised someone would write an article benchmarking a LLM use case without using the paid models initially.