An AI platform compares two models. One has cheaper tokens but needs more retries and manual correction. Which metric best reflects the business efficiency trade-off?

1. Only the number of model parameters.
2. Total cost per accepted successful task, including retries and review.
3. Only the lowest advertised input token rate.
4. Only the fastest single demonstration response.