Cursor Free Has Limits Nobody Will Put a Number On
The 2,000 completions and 50 requests figures come from users, not from Cursor. The vendor's own pricing page says limited and stops there.
Cursor's Hobby plan is free, needs no card, includes the full editor with limited tab completions and limited agent requests, and comes with a one-week Pro trial on signup. Cursor's pricing page does not publish a numeric quota for either limit. Pro is $20 a month, which buys unlimited tab completion and a usage pool metered at model API rates.1
Almost every guide states 2,000 completions and 50 slow requests. Those numbers are community reports from an earlier version of the plan. Cursor's current page describes both as limited and stops, which is why a careful review cannot tell you how much free usage you are buying.
That single fact should shape how you read any Cursor comparison, including this one. The editor is free and unrestricted. Only the AI features are capped, and the cap is unquantified.
What the free plan actually includes
| Hobby | Pro | |
|---|---|---|
| Price | $0 | $20/mo, $16 billed annually |
| Tab completions | Limited, no number | Unlimited |
| Agent requests | Limited, no number | Extended limits |
| Included model usage | None | About $20 of usage at API rates |
| Model choice | Auto only | Auto, Composer and frontier models |
| Cloud agents | No | Yes |
| MCP servers, skills, hooks | No | Yes |
| Bugbot code review | No | Usage-based |
| Editor features | Full VS Code compatible | Full |
| Trial | One week of Pro | Not applicable |
Two rows do more work than the rest. Model choice and MCP are absent entirely on Hobby, which means the free tier is not a small version of the product, it is the product without its integrations.
Pro is a dollar allowance, not a request count
Cursor moved Pro to usage-based pricing and the allowance is $20 of model spend at list prices, not a number of requests. Since August 2026 that is billed at the routed model's published API rate.
So a Pro user on a cheap model gets a lot of requests and one on a frontier model gets few, which is honest but means the plan cannot be compared to Copilot's 50 chat requests on the same axis. Copilot meters requests on Free and tokens on paid, and Cursor meters dollars on both.
Pro+ at $60 is roughly three times the pool and Ultra at $200 is about twenty times it, so the ladder is proportional rather than discounted.
What the free tier is genuinely for
The editor. Cursor is a VS Code fork and works fully without a subscription, including extensions, themes and every non-AI feature. Anyone who wants VS Code with a better default setup can use Cursor free indefinitely.
The one-week Pro trial is also the real evaluation path. Rather than trying to measure an unpublished free quota, use the seven days to see whether the agent fits how you work, then decide. That is precisely what the trial exists for.
Why the number is missing
Cursor has changed its packaging repeatedly, including a pricing reset in 2025 that replaced request counts with usage pools, and a cohort-by-cohort rollout where different accounts see different limits. Publishing a number that varies by account is worse than publishing none.
It sits at the opposite end of the spectrum from Copilot, which prints 2,000 and 50 and holds to them. Neither is obviously better. Copilot's number is useful and inflexible. Cursor's absence is flexible and useless for planning.
When the free plan is enough
Using the editor without AI. Trying agent mode during the trial to decide whether $20 is worth it. Light AI use on a small personal project where running out for a day is an inconvenience rather than a problem.
It is also a reasonable permanent home for a developer who already pays for a chat assistant elsewhere and wants completions occasionally. The absence of a published cap means you may get more than you expect.
When it stops being enough
The first time an agent request is refused mid-task, which happens without warning because there is no counter to check. That is the practical cost of an unpublished limit: you cannot ration what you cannot see.
The second is integrations. MCP servers, skills and hooks are Pro-only, and they are where Cursor's advantage over a plain editor lives. Free Cursor is a good editor with a small amount of AI attached.
At $20 against Copilot Pro at $10 and free Copilot's hard quotas, the decision is about whether you want a bounded amount of frontier model time or an unbounded amount of a cheaper one.
Frequently asked questions
Is Cursor free?
Yes. The Hobby plan costs nothing, requires no card, and includes the full editor with limited AI features and a one-week Pro trial.
How many free requests does Cursor give?
Cursor does not publish a number. Its pricing page describes agent requests and tab completions as limited. The widely quoted 2,000 and 50 figures are user reports, not vendor figures.
What happens when Hobby limits run out?
AI features stop until the monthly reset. The editor keeps working normally.
Does Cursor Free include frontier models?
No. Model choice is automatic on Hobby, and frontier models plus Composer require Pro.
Can I use my own API key on the free plan?
No. Bring-your-own-key requires Pro or higher.
Plan features, trial terms and pricing verified against Cursor's pricing page on 20 October 2026. Cursor publishes no numeric Hobby quota, and the figures circulating in third-party guides are community estimates rather than vendor statements.
Sources
- Cursor pricing: cursor.com/pricing. Used for: Plan pricing and the absence of a published free quota.
All sources are the vendor's own plan or documentation pages, checked on 20 October 2026. Plan terms change without notice, so confirm current limits before you rely on them.
