While working through a hard refactor on Auto, a developer wants one specific model to handle that single request. Which control does Cursor actually expose here?

The principle — Routing under Auto is decided by a classifier that inspects each request, not by the person sending it. Cursor exposes one dial over that process: the optimisation mode. Everything else about model selection is managed on Cursor's side, and the pool being routed across changes over time.

Why the key is correct — There is no per-request override, so no amount of searching the interface will surface one. What is available is the choice of optimisation mode, which shifts the whole distribution of models the router draws on. For a hard refactor the practical move is to raise the mode for that stretch of work, or to leave Auto entirely and select a model directly from the picker. Knowing this saves the time otherwise lost hunting for a control that was never built.

Why the others are wrong — Expecting a per-message pin comes from tools that offer one, but the point of Auto is that selection is automatic and data-driven. Naming a model in the prompt changes nothing about routing, because the classifier decides before the model sees the text, and that text is input to the model rather than configuration. A rules file cannot pin a model either; rules shape how the agent behaves once a model is handling the request.

Remember this — Inside Auto you steer with the mode. If you need a specific model, step outside Auto and pick it yourself.

Sources — Cursor's Cursor Router documentation.