[iliane5]

It’s not only 10x cheaper, it’s also way faster at inference and not as smart as Davinci. IMO the only logical answer is that the model is just smaller.