Auto Router [Add-on]

Auto Router chooses which model handles each request to your LiteLLM gateway. Your application calls one model name; the router selects from the models you configure based on the prompt or your routing rules. For example, you can send simple questions to a cheaper model and harder tasks to a stronger one.

Building routing into a Python application or agent? Use the Autorouter SDK.

You choose the models for each tier. For each request, the router picks a tier and a model from it.

Set up in the dashboard

You need a LiteLLM gateway with models already configured and permission to add models. If you do not have a gateway yet, start with the gateway quickstart.

Open Models + Endpoints, then the Auto Router tab, and add a router.
Give it a name, such as smart-router. Click Configure automatically to fill the tiers from your deployed models, then review the selections.
Use Test Routing with a prompt from your application. Check which tier and model it selects. This tests the routing decision without calling the selected model; an LLM classifier can still incur a charge.
Click Add Auto Router to save it. In your application's requests to the gateway, use smart-router as the model.

The selected model generates the answer. Try a few of your application's tasks and compare the answers and cost with your current model before switching traffic.

For YAML, the API, or CLI setup, see Admin Setup. To use the router from Claude Code or Codex, follow User Setup.

To change how the router chooses models, see Customize your classifier. You can also connect a decision model, including Jev, Nimble, Laya, or Databricks ai_decide.

Results

Result	Measured on	Read more
JEV: 5.43x as fast as Haiku by median classification latency and 96.12% lower registry-priced classifier cost	80 authored synthetic cases, three paired repeats, 95.00% vs 73.75% match with authored expected tiers under one fixed rubric	JEV comparison
Claude Opus-5 solve rate at 27% lower cost	21-task subset of Terminal-Bench 2.0, 16/21 solved by both	Terminal-Bench
Heuristic v2: 27% more tasks solved at 45% lower cost per task than v1	Same 21-task subset, no LLM classifier call	Heuristic v2
74.5% cheaper at 87.3% of frontier quality	RouterArena, 8,399 graded queries	Cost and quality
51.1% saved, $12,249 over four months	272,876 production requests, 450+ users	Production case study
37% to 69% cheaper than caching alone	Five datasets including live gateway traffic	Prompt caching
Matched or beat the current model on 88.1% of responses	Shadow evaluation on live traffic, 143 judged turns	Shadow evaluations