info snack
Model routing
Choosing the right model path for each request.
companion card
Model routing chooses which model handles a request based on cost, latency, quality, or policy.
What it means
A router sends requests to different models or deployments based on task type, customer, quality bar, fallback policy, budget, or live health.
Why product teams care
Easy requests can use cheaper routes, hard requests can use stronger routes, and sensitive requests can follow stricter policy or deployment rules.
Understudy angle
Understudy turns routing into an evidence problem: define workload, score candidates, and route only when a candidate proves it can serve.
take this with you
Routing lets one product use multiple models without paying frontier prices for every request.
Split one workload into easy, normal, and hard cases before choosing routes.