I've had the occasional large prompt that cost ~30c - I often use GPT-4o mini if I'm going to ask follow-up questions because then I get prompt caching without having to configure it.