is there any evidence that this is significant contribution?

My intuition that one need ALOT api credits to distill such large models.