A measure of how many tokens (small units of text) a model needs to use to complete a task; more efficient models use fewer tokens and cost less.