Comment by orangecat

Right, and it's often very sane. If you're paying $250k/year for a software engineer, it likely makes sense to have them spend $10k/year on tokens from the best available model rather than trying to save a few thousand with random Chinese models that may or may not be good enough.