The price of a request is decided after you send it.
Since 24 August 2026, every Cursor Auto request bills at whatever model it got routed to. The cheapest and priciest routed models are up to 20x apart for the same tokens, and a request routed to the priciest model runs about 8x what it would have cost under the old flat rate. Slide your real usage in and watch what the routing does to your month. Nothing leaves this page.
These sliders start on illustrative values, not a claim about a typical user. Edit them to match your Cursor usage dashboard, which is the only source that knows your real numbers.
Auto / agent requests only. Tab completions are not billed as routed Auto requests, leave them out. Your Cursor usage dashboard has the real count.
heavy = refactors, debugging, agent runs. We assume Auto sends heavy work 60/40 to Opus 5 / Fable 5, and routine work to Composer 2.5.
~2,000 tokens in / 500 out per request
Estimated routing premium for your assumed mix
more than pinning Composer 2.5, at 880 requests a month with 20% heavy work.
Theoretical maximum, if every request went to Fable 5: $37.6/mo. Where you actually land between $0 and that is a routing question your dashboard answers.
Honestly? At your volume the routing barely matters. Go build.
same requests. same you. three price tags.
estimates at list price, client-side only. your usage dashboard is the ground truth. prices verified against Cursor’s pricing docs on 2026-09-03.
You cannot control the prices. You can control the routing.
The headline is an estimate for the mix you assumed, and the theoretical maximum is the all-Fable ceiling above it. Which one you are near depends on where the router has actually been sending your requests, and that is knowable: twenty minutes in your usage dashboard settles it. The full walkthrough, including the two cases where switching actually pays, lives in our explainer: What does Cursor Auto actually cost per request now?
And if you want the routing decision back entirely: RelayPlane routes each task to the model you would have picked, with a local spend ledger. Free, open source, MIT.