Routers choose a model per request on price, speed or availability. CompletionPrism™ predicts the cost to finish the whole piece of work, rework included.
Routers and gateways choose a model on price, speed, availability or difficulty for a single request. CompletionPrism™ predicts the cost to finish the whole piece of work, including the human rework a wrong answer causes. It then chooses, before the request runs, whether to route down, route up, stay on the current model, or stop.
Picking the cheapest acceptable model is the default move of most routing tools. On hard work, that can raise the total cost. Read more in LLM Router vs AI Cost Optimizer, or see CompletionPrism™.