Hello! I have specific questions about how the toggles in the Model Settings interact, specifically about the “Premium” toggle. I know that the Premium toggle being on tells Cursor to decide which premium models to use and bills me accordingly. But I’m trying to understand:
When the Premium toggle is on, does Cursor use its own list and make its own decisions about model usage or does it consider which other toggles you have turned on? (i.e. Does it limit the list of models to what you have selected or does it ignore them?)
If the Premium toggle is off, and you have multiple other model toggles turned on, does this behave like a custom setup for Premium-like behavior (does Cursor use my list), or am I now expected to specify a config that explains which model is used for which task?
In general, how are these toggles expected to be used together? What are the best practices? What are the things to avoid doing?
I don’t know, I cynically assume it picks the cheapest (to Cursor) of the Premium models that its little prompt eval thinks can accomplish the job.
You specify the exact model you want to use, if Premium is toggled off, then you can either pick between Auto or a a specific model.
People will weigh in with different opinions here. I prefer always to specify which model I want to use for a given task. Sometimes that’s Composer 2, sometimes it’s Codex, sometimes it’s Sonnet, etc. I see no benefit in using the “auto-select” features at this point in time. As models become leaner, cheaper, and more specialized, this could change as I may need to decided between hundreds of models instead of just a few.
Premium toggle ON. Premium is a separate routing mode. Cursor picks the best premium model for your request. It doesn’t look at your per-model toggles and it isn’t limited by them, it uses its own internal list. Billing goes through the API pool at each model’s rates, not the Auto+Composer pool. See Models & Pricing | Cursor Docs, the usage pools section.
Premium toggle OFF + multiple per-model toggles ON. This isn’t custom Premium. Per-model toggles only control which models show up in the picker. Auto-routing within your list doesn’t turn on, you still pick the model in each chat. If you want auto selection, use Premium (API rates) or Auto (Auto+Composer pool, cheaper).
Best practices:
Auto is the default for daily work and it’s the cheapest.
Premium is for when you want Cursor to pick the strongest model for the request, and you’re ok paying API rates.
A specific model is best when you need predictable behavior or you know one model fits the task better, like Claude for refactors, GPT for certain kinds of reasoning.
Per-model toggles are just to keep the picker clean. Turn off the ones you don’t use.
What to avoid: don’t mix up Auto and Premium, they use different billing pools. And if Premium is on and selected in the picker, keep in mind you’re billed at API rates for whatever model it uses.