# Model routing

Directing a request to the most suitable model, provider, or serving path based on task requirements and economics.

The mechanism behind most real-world token-cost optimization: sending a simple request to a cheap, fast model and reserving an expensive frontier model for the requests that actually need it. Goodput-aware routing is the more disciplined version — it factors in the likelihood of a usable result, not just price.

**Related research:**
- [Token Economics: The Atomic Unit of AI Value](https://measuredvalue.ai/research/token-economics-atomic-unit-of-ai-value/)