Skip to main content
A model router picks which model handles each request. You send a router slug, such as openrouter/auto, as the model, and the router chooses a model from the catalog for that request. The response’s model field tells you which model served it. OpenRouter offers its own routers and also provides routers and composite models built by others. You can find them all in the model catalog. Model routing sits on top of provider routing. Once a router picks a model, OpenRouter chooses among the providers that host it using your provider preferences, live uptime and latency, price, and prompt cache state, the same way it does for a direct request to that model.

Choose a router

Auto Router

Auto Router and Auto Code run on the same routing engine. A fast decision model rates each request, and a routing policy weighs that rating against live prices, cache status, and benchmark results to pick a model and reasoning effort. Both work inside the model pool your guardrails, zero data retention requirements, provider restrictions, BYOK keys, and allowed and excluded model lists permit.

Specialty routers

Routing technologies

Compare routers on the Model Routing Benchmarks

The Model Routing Benchmarks page publishes score and cost per task for each router on OpenRouter, next to single-model baselines, across coding, agentic, and multimodal benchmarks. Runs repeat periodically.

Inspect routing decisions

Send the X-OpenRouter-Metadata: enabled header to get router metadata on the response. The router’s pipeline stage lists the models it considered, the one it picked, and why. In the chatroom, a routing insights panel shows the same analysis beside each reply.

Pricing

Routers don’t add a routing fee. You pay the standard price of the model that serves the request, and the context length is that model’s. Judge-backed Switchyard requests also bill a separate judge call.