I use GPT-4 through the API, but I can't help but hate the token/characterization pricing of these LLM APIs we've seen so far. Because the entire context needs to be fed back into the model, as my conversation gets longer, it gets more expensive. Yeah, it's fractions of a cent and cheaper, but something about it is so psychologically taxing that I'd rather pay a flat sum/month and get unlimited access, even if it costs more considering my usage.