Answers / Cursor Auto pricing

What does Cursor Auto actually cost per request now?

Since 24 August 2026, whatever the routed model charges. Auto's flat rate is gone, each request bills at the routed model's list price, and you learn the price after you send it. The bill is not the model, it is the routing, and the routing is fixable. Control which model each task goes to and the 8x surprise disappears. RelayPlane does exactly that, free and open source.

same request · routed model · list price /M tokens
composer 2.5$0.50 / $2.50
opus 5$5 / $25
fable 5$10 / $50
“the price of a request is determined after you send it.”

The short answer

Since 24 August 2026, Cursor Auto no longer has a flat rate. Each Auto request is routed to a model and billed at that model’s list price, so the same request can cost up to roughly 8x what it did before, and you only learn the price after you send it. To know your real per-request cost, open your usage dashboard, note which model Auto routed each request to, and multiply the tokens by that model’s list price. Switching pays in two cases: when Auto keeps routing routine work to expensive frontier models, or when your workload is predictable enough that a pinned cheaper model or a flat-rate tool does the same job.

The fix

You cannot control the prices. You can control the routing.

Getting squeezed by Auto is a routing problem wearing a pricing costume. Here is the playbook, in order.

1. Take the routing decision back. The 8x spread only hurts you because someone else decides which model runs your request. Route it yourself and routine edits go to cheap models while hard reasoning gets the expensive one on purpose. RelayPlane is the proxy I built in early 2026 when my own automated pipeline hit this exact problem, agents burning frontier-model money on routine work. It runs on your machine, routes each request by task complexity, and prices every call in a local ledger, with a hard cap so a runaway agent cannot run all night. It is free and MIT licensed, works with any model and any coding agent, and has logged 200,000 plus requests of my own production traffic since May.

2. Read your real bill once. Twenty minutes in the usage dashboard tells you which models Auto has been picking for you and what that actually cost. The four steps are below.

3. Then decide about switching. With your own numbers in hand, the two cases where leaving pays are easy to spot, and they are further down this page.

What changed

The flat rate died, the routing stayed.

On 24 August 2026 Cursor removed Auto's flat rate. Auto still does what it always did, it routes each request to whichever model it decides fits, but the request now bills at that model's list price. Cursor's own email called it a higher rate for most requests, the allowance increases were announced without numbers, and on Teams plans the routed model can be hidden from the person paying for it.

The spread is the story. A request routed to Composer 2.5 costs around $0.50 per million input tokens and $2.50 per million output. The same request routed to Opus 5 is $5 and $25. Routed to Fable 5, $10 and $50. That is how the same day of work can cost up to roughly 8x what it did the week before, with nothing about your workflow changing.

The complaint that sums it up: the price of a request is determined after you send it. If you are on a legacy Enterprise plan you keep the flat Auto rate until 7 September, which is a deadline, not a reprieve.

How to read your real bill

Four steps to your actual per-request cost.

Twenty minutes with the dashboard replaces a week of arguing about it on Reddit. Numbers first, decision second.

1
Open the usage dashboard
Look at individual requests, not the monthly total. The monthly number hides the routing decisions that are actually driving your bill.
2
Note the routed model per request
Every Auto request was sent to some model. That name is the price tag. If you are on a Teams plan and the routed model is hidden, ask your admin to surface it, you cannot manage a cost you cannot see.
3
Multiply tokens by that model’s list price
Input tokens at the input rate, output tokens at the output rate. The spread between a cheap routed model and a frontier one is roughly 8x, which is why two identical-feeling days can bill completely differently.
4
Compare a normal week against a pinned model
Run a week with routine work pinned to a model you chose, and compare. Now you know whether Auto’s routing is earning its convenience or just spending your money.
When switching pays

Two cases where leaving is the right math.

Case one: Auto routes routine work to expensive models. If your dashboard shows everyday edits and small refactors landing on frontier models, you are paying frontier prices for work a cheap model does fine. Pin a cheaper model for routine work, or move to a tool where the model choice is yours, and the bill drops the same day.

Case two: your usage is predictable. Variable pricing is a tax on predictability. If you use roughly the same models the same way every week, a flat-rate plan or direct API access to a model you chose is simply cheaper than list price on whatever Auto picks, and you stop being surprised by your own invoice.

If neither case matches your dashboard, the honest answer is that switching costs you a workflow and saves you very little. Being angry at a pricing email is not the same as being overcharged, and the twenty minutes of dashboard reading above settles which one you are.

A quieter question

Sometimes the bill is not the real problem.

A lot of the people reading pricing threads this week are not really deciding between editors. They are stuck. The app Cursor helped them build is 80 percent done, the auth kind of works, the payments pass the happy path, and nobody can say whether it is safe to put in front of strangers. Changing where the tokens are billed does not change any of that.

That last stretch is what I do. I am a senior engineer who takes AI-built apps the last mile, the security, the payments, the admin, the launch hardening, fixed scope for a fixed fee. If that is the situation behind your pricing frustration, start with the free Leak Check. It reads your live app from the outside, no code access, and tells you where you actually stand before you spend anything.

FAQ

Cursor Auto pricing, straight answers.

Cursor removed Auto’s flat rate. Auto still routes each request to a model automatically, but the request now bills at whatever that routed model charges at list price. Cursor’s own email described it as a higher rate for most requests, and the allowance increases that were announced came without numbers. Legacy Enterprise plans keep the old flat Auto rate until 7 September, then it ends for them too.
Because the price of an Auto request now depends on which model it was routed to, and the spread between models is large. A request routed to a cheap model like Composer 2.5 costs around $0.50 per million input tokens and $2.50 per million output tokens, while the same request routed to a frontier model like Fable 5 costs $10 and $50. That is the up-to-8x jump people are seeing, without changing anything about how they work.
Open your Cursor usage dashboard and look at the per-request entries. For each one, note which model Auto routed it to, then multiply the input and output tokens by that model’s list price. Do this for a normal day of work and you have your real per-request cost, not the average Cursor quotes you. On Teams plans the routed model can be hidden from you, which makes this harder, and is worth raising with your admin.
Two cases. First, if your dashboard shows Auto routinely routing everyday edits to expensive frontier models, pinning a cheaper model for routine work, or moving to a tool where you choose the model, cuts the bill immediately. Second, if your usage is steady and predictable, a flat-rate plan or direct API access to one known model is simply cheaper than paying list price for whatever Auto picks. If neither is true, the anger may cost you more than the bill does.
That is more common than people admit. If you are staring at the pricing page because the project itself is stuck, half-finished auth, payments that only work on the happy path, security you cannot vouch for, then switching editors will not fix it. That is the situation I work in every week: a senior engineer takes the AI-built app the last mile to a safe launch. The free Leak Check reads your live app from the outside and tells you where it actually stands, no code access needed.

The bill is not the model.
It is the routing.

Stop paying for the wrong model →Book a call
RelayPlane routes each task to the right model and shows your real per-task spend. Free and open source.30 minutes. Tell me where the project is stuck and I will tell you what finishing it looks like.

If the honest answer is that your bill is fine and your app is fine, that is the answer you will get.

Matt