Automatic model selection with Router

Simple tasks rarely need the most expensive model. Router selects models by task complexity to manage cost and latency.

Router uses prompt length, tool calls, images, and session depth to classify requests as trivial, standard, or complex. Tool use and file edits require at least the standard category.

Router uses a fast classifier to check requests within 5% of a category boundary.

Router records the model, task category, selection source, latency, and cost so you can inspect its choices and spending.

If a provider rate-limits the selected model, Router falls back to another model allowed by your plan.

Router keeps the current model when it can handle the task, preserving cached context and avoiding repeat token costs.

You can pin a model to override Router for the session.

Choose Cost to favor cheaper models or Intelligence to use more capable models sooner. Router uses Balance by default.

Router applies your preference in terminal, web, and cloud sessions.