You need to compare cost per task buddy boy. Cost per token doesn't tell you much when you don't know how many tokens a model will use to accomplish a task