Or you could use a local model where you’re not constrained by tokens. Like rig.ai