Looks like good one this time. Moving from fine tuned gpt-4.1-mini for structured outputs to gpt-5-mini made zero sense just because 5 was reasoning model and there was no way to disable the reasoning and it also did not have support for fine-tuning.
So essentially I was not able to get nowhere close to the accuracy of previous model and it was slower, and more expensive at the same time.
Now gpt-6-luna, has really competitive pricing and offers similar accuracy compared to gpt-4.1-mini fine tuned for my specific task. And fine tuned models are getting deprecated anyways, seems like a good time to move to gpt-6-luna.
So essentially I was not able to get nowhere close to the accuracy of previous model and it was slower, and more expensive at the same time.
Now gpt-6-luna, has really competitive pricing and offers similar accuracy compared to gpt-4.1-mini fine tuned for my specific task. And fine tuned models are getting deprecated anyways, seems like a good time to move to gpt-6-luna.