Model routing dispatches each request to the most appropriate model, cheap fast models for routine tasks, frontier or reasoning models for hard ones, based on t
Dispatching each request to the most appropriate model, cheap fast models for routine tasks, frontier or reasoning models for hard ones, based on task type, complexity signals, cost budgets, and fallback logic when providers fail.
Because most traffic is routine, and sending it all to premium models is the classic budget leak. Routers live in gateways, with eval suites verifying that downgraded traffic does not lose quality.