AI-103: Developing AI Apps and Agents on Azure

A model cascade (model routing) evaluates each request and forwards simple queries to a small, fast, inexpensive model while escalating requests that need advanced reasoning to a larger, more capable model, so common FAQ traffic is served cheaply and quickly while complex questions still reach a capable model, preserving their answer quality. Routing every request to a smaller model would degrade the complex answers, routing everything to the most capable model keeps cost and latency high even for simple queries, and increasing max_tokens changes how much output a response can contain, not which model handles the request or its cost and latency profile.