Why per-token AI API billing charges for effort rather than results

An analysis argues that billing AI APIs per token misaligns cost with value, since users pay for every token a model generates, including failed calls, retries and unseen work. The author compares it to a taxi meter that charges for wheel turns rather than arrival. Reasoning models illustrate the problem: hidden reasoning tokens are billed at the output rate even though users cannot read them. Raising a reasoning-effort setting can make the same correct answer cost more, showing the meter tracks process rather than value. The proposed alternative is pricing per completed task instead of per token spent.