Skip to content
ML Academy

Introduced on day 50

Cost per supported answer

Total spend divided by the number of answers that were correct and supported by the evidence, which includes the cost of requests that did not produce one.

In detail

Price per request understates the unit cost of useful output because failed and abstained requests are paid for as well. Cost per supported answer falls when recall rises, when retries and padded contexts fall, and when routing sends easy requests to cheaper models. Prices used in a review should be labelled as assumptions.