The true cost of Cursor Pro now that it bills at API rates

Cursor Pro is still $20 a month. What that $20 buys has changed, and most of the cost math you will find online -- including the first version of this post -- is built on the old model.

What changed in this post

This post used to count "fast premium requests": 500 a month, then $0.04 each, and a projection that a heavy user paid $60 a month. Cursor no longer sells requests that way. It sells a pool of included usage, metered at each model's API price. The numbers below come from Cursor's pricing documentation as of October 2026, and the projection is gone, because a request count no longer predicts the bill.

What the plans are now

Cursor's docs describe two included pools, one for Cursor's own models and one for other providers' models, and state that "your model selection affects how quickly your included usage is consumed." Auto mode bills at the list price of whichever model it routes each request to. When the included usage runs out, you can turn on on-demand usage, which continues at the same API rates and is billed in arrears, or move up a plan.

What that means for your bill

The question is no longer how many requests you make. It is which models you use and how much context each request carries.

• Expensive models drain the pool faster. The same agent session costs very different amounts on a frontier model and on a small one.
• Long agent sessions cost more per step. Every step resends the conversation so far, so the late steps of a long session are the expensive ones.
• Overage is not a penalty rate. On-demand usage is charged at the same API prices, so going over costs what the extra tokens cost, not a surcharge.

The practical move is to watch the usage page in your Cursor dashboard for a week or two before deciding anything. That tells you more than any estimate here can.

How the alternatives bill now

The other subscriptions moved the same way:

• Devin Desktop (formerly Windsurf). Cognition renamed Windsurf in June 2026. Pro is $20 a month and Max $200, with an allowance that refreshes daily and weekly, and usage beyond it charged at API pricing.
• GitHub Copilot. Usage is metered in AI credits worth $0.01 each. Pro is $10 a month, Pro+ $39 and Max $100, each with a monthly credit allowance.

None of them is flat-rate any more. Comparing them on price alone mostly comes down to how much included usage each plan gives for your mix of models.

Where your own keys fit

Cursor lets you add your own API key for chat models. Its docs say that on individual plans "your provider bills you for the model cost, and those requests don't draw from your included usage." Tab completion keeps using Cursor's own models either way.

That is the opening for a router. Point Cursor's custom OpenAI key and base URL at kRouter and the chat requests go to whatever kRouter has behind it: a free tier such as Kiro's 50 credits a month, a GitHub Copilot subscription you already pay for, or a cheap metered API, in a fallback combo:

1. kr/claude-sonnet-4.5    # Kiro free tier
2. gh/claude-sonnet-4.6    # Copilot subscription
3. glm/glm-5.3             # metered overflow

Three caveats before you rely on it:

• localhost does not work. Cursor sends these requests from its own servers, so they cannot reach a router on your laptop. You need kRouter's Tunnel or Cloud endpoint, and because those requests arrive from outside your machine, an API key from kRouter's Endpoint page. The Cursor localhost guide walks through it, and kRouter's own Cursor guide notes that the feature requires a Cursor Pro account.
• Only chat moves. Tab, and anything Cursor runs on its own models, still draws from your plan.
• Subscription logins carry account risk. Kiro and Copilot sessions are licensed for their own clients. kRouter shows a risk notice before you connect them, because the provider can restrict an account it sees used through a proxy.

When to skip Pro

If you mostly use Tab and the occasional chat, Hobby may be enough. If you do heavy agent work on frontier models, the included pool on Pro will not cover it whatever you do, and the real choice is between a bigger Cursor plan, on-demand usage at API rates, or routing chat to access you already pay for.

npm install -g @sifxprime/krouter

Common questions

Does Cursor Pro still include 500 fast requests?

No. Cursor replaced the request count with included usage metered at each model's API price. Which models you use, and how much context each request carries, decides how fast it runs out.

What happens when my included Cursor usage runs out?

You can enable on-demand usage, which continues at the same API rates and is billed in arrears, or upgrade to a plan with more included usage.

Do requests with my own API key count against Cursor's included usage?

On individual plans, no: Cursor's docs say your provider bills you and those requests do not draw from your included usage. Tab completion still uses Cursor's own models.

Can I point Cursor at a router on localhost?

Not directly. Cursor calls custom endpoints from its own servers, so the router needs a public URL, such as kRouter's tunnel, and an API key.

Related

• Cursor usage limit reached: what to do
• Best AI coding assistant in 2026: how to choose
• Windsurf vs Cursor in 2026