If cost is your primary concern, shouldn't you support ByteDance's efforts to reduce inference costs by distilling the model?

(while at the same time reducing future costs for everyone by distributing the capability more widely to prevent monopolization)