It looks like it did give me the $18 when I first created a chatGPT account

OpenAI's GPT-4o-mini model is significantly cheaper than previous models - they also claim it does a better job which seems to be correct in the use cases we put it to. However, being non-deterministic, it's quite difficult to objectively test...

I have an AI language partner that has conversations with me in Turkish and corrects any major mistakes I make. We also have a summary of the needs of a dog coming into our care for my partner's business Pawsies. In both of these cases, GPT-4o-mini produces better results.

Today I've released an updated version of AI::Chat to CPAN that uses GPT-4o-mini by default instead of GPT-3.5-Turbo-0125.

Whether or not it's better, it's definitely cheaper. Within On Radar, our customers are doing more things that result in an API call, but our costs are decreasing.