Comment by terekhindc

the per-token comparison keeps missing that k3 spends way more tokens per task. if it burns 3x tokens to reach the same result as fable, cheap per-token stops mattering