It is listed as a model on the usage page. And not like its balanced or smart veriant, the usage page does not show what model it truely is e.g. when routed to other models the usage page will show something like GPT-5.6 Sol (Auto Balanced)
The question is is Auto a model? And if so, how is it charged?
Hey @NoMames,
Auto isn’t a model, it’s a routing option: each request is handed to one of the Cursor models based on the task, and it’s billed at the list price of whichever model actually handled it (no separate “Auto rate”) - rates on the Models & Pricing page.
The Usage page just labels these rows “Auto” and doesn’t yet show the routed model on non-enterprise plans, whereas the Router modes show “ (Auto Balanced)”. If you want the model per request now, pick one directly (e.g. Composer 2.5 or Grok 4.6) and it shows by name.
So it’s basically composer 2.5 when doing easy tasks and grok 4.6 when doing something more complex?
Roughly, yes. Auto routes among Cursor’s own models, mainly Composer 2.5 and Grok 4.5 / 4.6. Lighter work tends toward Composer and heavier work toward Grok, but it’s decided per request rather than a fixed rule, and the exact lineup can change over time. Each request is billed at the list price of whichever model handled it.