Source: OpenRouter Auto Router docs
Describes openrouter/auto, a hosted per-request model router that rates the request with a fast decision model and picks a model and reasoning effort within a cost tier.
Contents
- What: official documentation (“Model Routing” / Auto Router)
- Publisher: OpenRouter
- Date: living docs, read 2026-11-14
- Original: openrouter.ai/docs/features/model-routing
Summary
openrouter/auto is a general-purpose router exposed as a model slug. It filters the candidate pool by policy, rates the request with a fast decision model, and chooses a model and reasoning effort using live prices, cache status and benchmark data.
Key claims
- Pool narrowing first: guardrails, zero data retention, capabilities, provider restrictions, BYOK keys, allow/deny lists.
- A fast decision model rates eleven aspects of the request (task type, difficulty, need for vision, etc.) in one call.
- The policy weighs that rating against live prices, prompt-cache status and benchmark performance within the caller’s cost tier; then normal provider routing applies.
- Within a session it switches models only when the gain beats the cost of rebuilding the prompt cache, and may change reasoning effort instead.
- No extra fee: the caller pays the selected model’s rate;
provider.max_pricestill applies. Variants:openrouter/auto-code(coding agents) andopenrouter/auto-beta.