The new metric for AI efficiency isn’t price per token on your model, but cost per task. This gets tricky, as the best model for the job isn’t always apparent. The best solution for most business users, focus on the harness first, where big savings are usually available.