Sending each request to the cheapest model that can handle it.
1 min readupdated 2026-07-04
/ quick answer
A router classifies the task (simple/hard, short/long, code/prose) and dispatches to a small model (Haiku, GPT-4o mini) or a large one (Sonnet, GPT-4o). Cuts cost 5-10x with no visible quality loss. Sending each request to the cheapest model that can handle it.
Sending each request to the cheapest model that can handle it. A router classifies the task (simple/hard, short/long, code/prose) and dispatches to a small model (Haiku, GPT-4o mini) or a large one (Sonnet, GPT-4o). Cuts cost 5-10x with no visible quality loss. In practice: A support bot routes FAQ matches to a small model and only escalates ambiguous queries to Sonnet. This dictionary node is part of the Onexial knowledge graph and links to related concepts, workflows and tools below.
Definition
A router classifies the task (simple/hard, short/long, code/prose) and dispatches to a small model (Haiku, GPT-4o mini) or a large one (Sonnet, GPT-4o). Cuts cost 5-10x with no visible quality loss.
Example
A support bot routes FAQ matches to a small model and only escalates ambiguous queries to Sonnet.
A router classifies the task (simple/hard, short/long, code/prose) and dispatches to a small model (Haiku, GPT-4o mini) or a large one (Sonnet, GPT-4o). Cuts cost 5-10x with no visible quality loss.
What is an example of Model Routing?
A support bot routes FAQ matches to a small model and only escalates ambiguous queries to Sonnet.
Why does Model Routing matter for AI and automation?
Sending each request to the cheapest model that can handle it. It connects to the workflows, prompts and tool stacks linked on this page, so you can move from definition to execution without leaving Onexial.