i think tokenmaxxing mostly comes from cloud pricing. once you're paying by the token, you naturally start caring about token counts. with local inference i barely think about it anymore.