[theRealMe]

I don’t want to make unsubstantiated guesses, but we should remember that GPT3.5 was originally just GPT3.5 and when they came out with turbo it not only sped it up but also reduced the cost (to us) by like 10x of what it used to be if I remember right. So they are (or at the very least “were”) working on and succeeding on reducing its cost to run.