---
title: "LLM Routing (RouteLLM) · agent orchestration · Agent Fieldbook"
url: https://agentfieldbook.org/design/orchestration/routellm/
description: "A learned router scores each query and sends easy ones to a cheap model, hard ones to a strong model. Trained on human-preference signals to hold quality while cutting spend."
section: "Design · Orchestration"
source: Agent Fieldbook — generated from the published page
---

# LLM Routing (RouteLLM)

A learned router scores each query and sends easy ones to a cheap model, hard ones to a strong model. Trained on human-preference signals to hold quality while cutting spend.

Verified against [arxiv.org](https://arxiv.org/abs/2406.18665)

- [arxiv.org](https://arxiv.org/abs/2406.18665)

## How it works

## At a glance

| Field | Value |
| --- | --- |
| Technique | Routing |
| Cost lever | Much lower, up to 85% cost cut at GPT-4-level quality on MT-Bench |
| Latency lever | Lower on average, most queries hit the small model |
| Quality lever | Near-frontier when routed well |
| Example | LMSYS RouteLLM· Martian· OpenRouter auto |

## When to use

High-volume mixed-difficulty traffic where most queries are easy and a minority need a frontier model.

## Basis

## Verified against

- [https://arxiv.org/abs/2406.18665](https://arxiv.org/abs/2406.18665)

The corpus does not rewrite vendor documentation. Where copy is quoted verbatim it is marked as such and attributed to its source.