As the other reply brought up, GPT-4 access is waitlisted, and costs significantly more API access makes it easy to chew through a lot of tokens. With GPT3.5 I

As the other reply brought up, GPT-4 access is waitlisted, and costs significantly more API access makes it easy to chew through a lot of tokens. With GPT3.5 I hardly have to think about the cost of tinkering for hours on end. Yesterday I made over 600 requests and wracked up a whopping 67 cent bill... At an order of magnitude more, I'd still tinker, but wouldn't feel comfortable with all of the same usecases (ie. setting it up in ways that allow it to recurse very quickly like I do with 3.5)