Hey Cursor team,
I enabled Cursor Router for our org and a question came up about context limits.
When selecting a model manually, we can choose a smaller or larger context window depending on the model. With Router enabled, that option isn’t there.
Does Router always use the smaller context window or can it also choose a larger one based on the task?
If it switches automatically, what determines when that happens?
Hi @janjic-ps Thanks for the post!
Yes, Router always uses each model’s standard context window. It never selects the extended 1M option, and task size or conversation length doesn’t change that.
The exact size depends on which model Router picks for a given request. With the current model pool, that’s 256K to 300K tokens. If you specifically need 1M context, select a supported model manually. For longer Router conversations, Cursor summarizes earlier context as the window fills.
For more details: Cursor Router | Cursor Docs