One thing that was complex to figure out was that many models have different pricing structures, so defining request costs is not easy. Neither fal nor replicate return accurate pricing in the response body.

So I have a script running every hour through my fal requests and compares the prices I paid to the ones I've charged and alerts me if there are discrepancies. I'd appreciate suggestions on how to improve this part though