[FloorEgg]

This was exactly my experience. I have very complex prompts and I test them on new models and nothing performs as well as GPT-4 that I've tried (Claude 3 Opus included)