Cursor Pricing in 2026: Free, Pro, Business, and Ultra Explained
Cursor is one of those tools where the pricing page alone doesn't really tell you what you need to know. The plan names are simple enough: Free, Pro, Business, Ultra. But whether any given plan is worth it depends on how you actually code, how much context you push through, and whether you're on a team or flying solo.
Let me walk through each tier in real terms, with the math that the marketing page glosses over.
Free tier: actually usable, with real limits
Cursor Free gives you 2,000 completions per month and 50 "slow" requests to premium models. That second number is the one that bites people. The 50 slow requests means 50 turns with GPT-4 or Claude-class models, processed in a lower-priority queue. After that, you're on cursor-small, Cursor's faster but weaker in-house model.
For occasional hobbyist use or evaluation, this is fine. You'll hit the ceiling fast if you're doing anything serious. Two weeks of active coding and you'll be refreshing the request counter.
The completions (tab completions, the inline autocomplete) are 2,000 per month, which sounds like a lot until you realize a focused coding day can burn through 100-200 of them. That's roughly 10-20 working days before you run out, assuming average usage. Heavy users, or anyone doing exploratory coding with lots of back-and-forth, will exhaust this in under two weeks.
One useful detail: requests to weaker models (cursor-small, GPT-3.5-class) don't count toward your premium quota. If you're willing to use those for simpler tasks, you can stretch the free tier meaningfully.
The verdict on Free: it's good for trying Cursor and decent for light weekend projects. It's not a plan you can run a real development workflow on.
Pro plan: $20/month, the one most developers end up on
Cursor Pro is $20/month (or $16/month billed annually at $192/year). This is the core product for individual developers.
What you get:
- 500 "fast" premium model requests per month
- Unlimited slow requests
- Unlimited completions
The completions being unlimited changes the math significantly. Tab completions are Cursor's most-used feature by usage volume, and removing that cap makes the product feel different from Free.
The 500 fast requests is where it gets interesting. A fast request is a real-time response from a frontier model, currently GPT-4o, Claude 3.5 Sonnet, or Claude 3.7 Sonnet depending on your settings and availability. These are the responses you get during active Chat sessions, when you're asking Cursor to refactor a function, explain a bug, or write something new.
500 requests sounds generous, but if you're using Chat heavily and working through complex problems, 500 can evaporate in a couple of weeks. An intensive session debugging a gnarly issue might burn 30-40 requests in an afternoon. Multiply that across a month and you can see how it works out.
After you exhaust your 500 fast requests, Cursor falls back to slow requests, which still use the same models but are queued lower priority. In practice, slow requests add a few seconds to response time during peak hours. For most non-real-time uses (refactoring, code review, writing) the quality is identical, just slower. You won't notice it at 2am; you will notice it at 2pm on a Tuesday.
What Pro doesn't include
Pro is a single-user plan. No team seats, no centralized billing, no admin controls. If you're managing multiple developers, you'll be chasing individual subscriptions, which gets messy fast.
Pro also doesn't include extended context modes that some power users want for large codebases. You're working within the standard context window the models provide, which for Claude 3.7 is 200K tokens, plenty for most sessions but not unlimited.
Business plan: $40/seat/month, team features kick in
Cursor Business is $40/seat/month billed annually. There's no monthly billing option for Business, so you're committing to $480/seat/year.
What Business adds over Pro:
- Centralized team billing and invoicing
- Admin dashboard for managing seats and access
- Usage analytics across the team
- Privacy mode enforced organization-wide
- SSO (SAML/OIDC) for enterprise identity management
- Zero data retention: Cursor doesn't log your code prompts
That last point is significant for teams working on proprietary code. On Free and Pro, Cursor's default settings allow some data to be used for product improvement. On Business, you get hard contractual guarantees that your code isn't stored or used for training. For agencies, consulting firms, or companies with sensitive codebases, that's often the deciding factor.
The model access on Business is the same as Pro in terms of which models you can access. You don't get more fast requests per seat on Business. What you get is administrative control and the privacy guarantees, not a raw capability upgrade.
The pricing math for teams
At $40/seat, Cursor Business costs twice what Pro costs per person. For a five-person team, that's $2,400/year versus $960/year for five individual Pro subscriptions.
What you're paying for: one invoice, one admin panel, enforced privacy settings, and SSO. Whether that's worth the premium depends entirely on your company's security posture and how much you care about centralized management.
Small teams where everyone's comfortable with individual billing often stick with Pro. Teams at companies with IT policies, vendor reviews, or compliance requirements usually have no choice but to go Business.
Ultra plan: $200/month, for power users who've done the math
Cursor Ultra launched in late 2025 and targets a specific audience: developers who regularly hit Pro's request limits.
Ultra costs $200/month with no annual discount option as of early 2026. That's $2,400/year, 10x the Pro annual price.
What you actually get:
- 10x more "fast" request quota than Pro (so approximately 5,000/month)
- Priority access during high-demand periods
- Access to the most capable model variants as they release
Ultra doesn't remove the model caps entirely. You're still working with tokens per context window; you just have more turns at the table each month. And you get priority queuing, so you're not waiting during peak hours.
Who actually needs Ultra
The honest answer is: very few people. 500 fast requests per month covers most professional developers comfortably. You need to be doing intensive, sustained AI-assisted coding across long sessions, multiple days a week, to reliably exhaust 500 requests.
The users who make Ultra make sense:
- Full-time AI-assisted developers where Cursor is open all day, every day
- Developers working on multiple large projects simultaneously
- Teams doing heavy code review through Cursor Chat
- Anyone who's hit Pro limits three months in a row and knows it'll keep happening
If you've never gone over 500 fast requests in a month, Ultra is spending $180 more for headroom you won't use. Cursor shows your usage in the settings panel, so it's easy to check whether you're consistently hitting the ceiling before committing.
Token math: what actually consumes your requests
Understanding what burns through requests helps you use any plan more efficiently.
Each "fast request" in Cursor's counting is roughly one turn in a Chat session: you send a message, the model responds. Complex multi-step tasks that Cursor handles automatically (like Composer working through multiple files) can count as multiple requests behind the scenes.
Some ways to stretch your quota:
- Use tab completions (the inline autocomplete) for simple additions. These don't use your premium request quota.
- Switch to slow requests for non-urgent work. Same model quality, just queued.
- Use cursor-small for quick, simple questions where a weaker model is fine.
- Keep context tight. Unnecessarily large context windows don't increase your request count, but they do slow things down and reduce quality.
The token math inside each request matters too. Cursor sends your recent files, the surrounding code, your chat history, and system context along with every message. A session with 10 open large files costs more tokens per request than a focused session on one file. This doesn't change your request count, but it does affect how much context Cursor can actually use.
Comparing the plans directly
| Feature | Free | Pro | Business | Ultra |
|---|---|---|---|---|
| Price | $0 | $20/mo | $40/seat/mo | $200/mo |
| Completions | 2,000/mo | Unlimited | Unlimited | Unlimited |
| Fast requests | 50/mo | 500/mo | 500/seat/mo | ~5,000/mo |
| Team billing | No | No | Yes | No |
| Privacy mode | Optional | Optional | Enforced | Optional |
| SSO | No | No | Yes | No |
The real decision tree
Start with Free if you want to evaluate Cursor without spending anything. Give it a real two weeks of usage to see how you actually interact with it.
Move to Pro when you're hitting the Free limits regularly and you're working on real projects. The $20/month is a straightforward call for most employed developers.
Go Business when your company has compliance requirements, needs centralized billing, or wants SSO. The privacy guarantees alone are often worth it for commercial work.
Consider Ultra if you've been on Pro for at least three months and you've consistently exhausted your fast request quota. Don't buy it speculatively.
One thing worth knowing: Cursor's pricing has evolved, there were adjustments in 2025, and they may change again. The request limits in this article reflect the April 2026 plan structure. Before committing to annual billing, check the current plan page to confirm the numbers haven't shifted.
For code-focused teams also evaluating GitHub Copilot, the comparison between Cursor and Copilot covers how the two products differ beyond just pricing.