#cost

Question 1: Single choice

An agent service routes simple requests to a small, cheap model and reserves a large model for hard reasoning. What should decide which request types the small model may handle?

Its lower price, which makes it right for every request that is not obviously hard1
The small model's own confidence in each answer2
The length of the user's message3
An evaluation showing that the small model is good enough on those request types4

Timed test