Not all prompts are created equal. You can save a bundle on token costs by routing your simpler prompts to cheaper models.