Cursor Pricing Explained
Cursor's advertised monthly fee buys an editor plan with included AI usage, not an infinite number of frontier-model tokens. Prices and pools can change; check the linked official page before buying.
An independent Licuse guide. We link to product documentation so you can verify current features and pricing.
Current individual tiers
As checked in September 2026, Cursor lists Hobby as free, Pro at $20/month, Pro+ at $60/month and Ultra at $200/month on its official pricing documentation. Annual offers and taxes can change the amount you pay.
Usage pools
Cursor describes separate pools for its own models and third-party models. Model choice, prompt length and output all affect consumption. On-demand usage can continue after the included amount if enabled, so inspect spend settings.
BYOK is a separate path
Cursor supports custom provider keys. Its current docs say individual BYOK requests do not draw down included usage, but the provider bills the model. Team and enterprise routing charges have separate rules.
The plan fee is only the first line
Cursor's advertised plan price pays for access to an editor and a defined set of AI allowances. It is tempting to divide that fee by an estimated number of requests, but agent requests vary widely. A short explanation and a repository-wide repair can use very different models, context lengths and tool loops. The included pool, on-demand settings and any provider-key route determine the rest of the bill. Start with the live pricing page and your account dashboard, because model pools and promotions can change. Then separate fixed subscription cost from variable model use. If your organization pays for seats and API keys, account for both. A free plan can be useful for evaluation, but its limitations may make it a poor proxy for a month of production work.
Understand the included pool
Cursor documents different allowances for its models and third-party models. The pool is not a simple entitlement to a fixed number of coding tasks. More context, a costly model or repeated agent steps draw down allowance faster. Some work may be covered by a separate category, so inspect the current dashboard rather than extrapolating from one request. A practical check is to run three representative tasks: a small question, a multi-file edit and a longer investigation. Note the displayed usage after each. Repeat with your preferred model. This turns an abstract plan label into an estimate tied to your repository. If the trial uses a different model from daily work, its cost forecast will be misleading.
What happens at the boundary
A plan can feel inexpensive until the included amount is exhausted. Depending on account settings, you may switch models, wait for a reset, upgrade or use on-demand billing. On-demand charges can be reasonable when occasional heavy tasks are important, but they should be deliberate. Review the spend controls available in your account and decide an upper monthly bound before enabling overage. Team administrators should check whether member-level settings and shared billing behave as expected. A notification after the fact is less useful than a budget that prevents surprise. Always verify the current rules in Cursor's documentation; pricing mechanics are product settings, not timeless properties of the model.
BYOK changes who bills inference
Cursor supports custom provider keys for supported model use. Its documentation says individual BYOK requests do not draw from the included model pool, while provider usage is charged to the key owner. That does not mean the whole Cursor experience becomes free or that every feature uses the custom key. Cursor also says requests pass through its servers to assemble the final prompt. If direct provider routing is a requirement, this distinction matters alongside price. A company key may have its own quotas and retention policy. Give it a limited budget, test one model and inspect provider usage alongside Cursor's dashboard. Do not assume the two dashboards count the same event in the same way.
Compare with alternatives fairly
A Cursor subscription can be a good deal for someone who uses its editor daily and benefits from the included usage. A BYOK editor may have a lower fixed fee but higher variable model charges. A CLI agent may be covered by another subscription while requiring a separate editor. The relevant unit is a verified task, not a generated line or a chat turn. Record the full month: plan fee, overage, provider charges and any additional tools. Include human repair time when an option repeatedly produces weak patches. The cheapest bill can be the most expensive workflow if it creates extra review work. Conversely, a higher fixed fee may be unnecessary for light or intermittent use.
An example budget exercise
Imagine a developer who mostly asks short codebase questions but completes two substantial agent tasks each week. On the first day of a billing cycle, record the starting allowance. After a week, inspect how the tasks drew from each category and whether any on-demand charge appeared. Estimate a month from that sample, then leave margin for unusual work. Do not multiply a single expensive migration by four if it was exceptional. The point is to build a forecast from your own model mix and repository size. Recheck the forecast after changing plans, models or context settings because all three can alter consumption.
When to choose a tier
Start at the lowest tier that supports your required workflow. Upgrade when measured usage or a necessary feature justifies the higher fixed fee. Downgrade if the included pool routinely remains unused. For a team, evaluate seats with different workloads rather than assigning the power-user tier to everyone. A good pricing decision is reversible and based on observed use. Keep the official pricing and model pages close to the decision, because a comparison written today can become outdated after a product change. The account dashboard is the final source for what your current settings will actually charge.
Read the invoice, not just the badge
When a bill differs from expectation, trace it to the active model and feature. A premium model selected for an agent session may account for more than dozens of small questions. Check whether on-demand usage was enabled and whether a provider key was in use. If a shared team account pays the provider invoice, attribution may require separate projects or keys. This investigation should be done before changing tiers, because an accidental route or runaway session is not evidence that every developer needs a more expensive plan.
Treat feature value separately
Some developers value Cursor's integrated navigation and agent workflow enough to pay a plan fee even when BYOK inference is billed elsewhere. Others mainly want occasional model access and may prefer a lower fixed commitment. Keep feature value separate from raw token economics. Ask whether the paid tier saves setup or review time in your actual repository. This prevents a cost comparison from undervaluing useful editor features or from treating unused allowance as a benefit simply because it appears in a plan description.
Compare your expected model mix and overage settings with the live Cursor dashboard before choosing a tier.
Sources & further reading
Product features and prices can change. We checked these official pages on September 26, 2026.
One workspace for code and notes.
Explore the beta, connect a supported provider key and keep the work in view.