A model router chooses the model for each request instead of hardcoding one. It can weigh cost, required capabilities (tools, vision, JSON), context size, provider health, and quality. Routing is the biggest lever for cutting spend without changing application behavior.
Why it matters
Hardcoding one model forces a single price and quality point for every request. Routing adapts the model to the request, which is the biggest lever for cutting spend without changing behavior.
How it works
The router scores candidate models against the request - required capabilities, context size, cost, provider health, and quality - then picks one. Simple requests go to cheap models; complex ones go to stronger models.