(After a day of usage, I am relatively certain in practice this does not end up being a 5.7x cost increase or anything close to that, though I am still fairly unclear on what that computation is worth to begin with, given that I am entirely fine with the model using the least amount of tokens possible to get the job done)