Cursor has five pricing tiers in 2026: Hobby (free), Pro ($20/month), Pro+ ($60/month), Ultra ($200/month), and Teams — which now splits into Standard ($40/user/month) and Premium ($120/user/month) after a June 2026 update. Most professional developers land on Pro. Most teams land on Standard.
The credit system changed in mid-2025 from fixed request counts to a dollar-value pool. Cursor Pro includes $20 in monthly model credits — enough for roughly 225 Claude Sonnet requests or 550 Gemini requests. Auto mode, Cursor's built-in model router, is genuinely unlimited and does not draw from that pool. You only consume credits when you manually pin a specific premium model by name.
Hobby: what the free plan actually includes
Cursor's free tier is called Hobby and is available permanently with no credit card required. New accounts automatically start a 2-week Pro trial — you get full Pro capabilities for two weeks before reverting.
Hobby includes limited Agent requests per month, limited Tab completions, and access to the core editor. No cloud agents, no background agents, no priority access to new models.
The 2-week trial makes a stronger first impression than GitHub Copilot's permanent free tier, which allows 2,000 completions and 50 chat messages per month. But once the trial ends, Cursor's Hobby fallback is tighter than Copilot's free plan. If you mainly want autocomplete and want to evaluate without commitment, Copilot's free tier gives you more room.
The honest test for whether you should pay: if you miss agent mode on the first day after the trial ends, buy Pro. If you don't notice the downgrade, Hobby is enough.
Pro: $20/month and how the credit pool works
Pro costs $20/month, or $192/year with annual billing — a 20% saving worth taking once you are past the evaluation stage.
What Pro includes: unlimited Tab completions, extended Agent limits, access to all frontier models (Claude 4.x, GPT-4o, Gemini 2.5), MCPs, skills, hooks, cloud agents, and Bugbot on usage-based billing. The defining item is the $20 monthly model credit pool.
How the credit pool works: $20 buys different amounts depending on the model. Manually selecting Claude Sonnet 4 gets you roughly 225 requests before the pool runs out. GPT-4.1 gets you roughly 650. Gemini models get roughly 550. These are not fixed caps — they vary by how much context you include per request. A long multi-file refactor costs more than a quick autocomplete.
Auto mode is the important exception. When you use Auto — Cursor's default router that picks the best model for reliability and cost — usage does not draw from the $20 credit pool at all. It runs at flat usage-based rates with no hard ceiling. Most developers who let Cursor route models find they rarely exhaust their credits.
When credits run out, on-demand billing starts automatically at token-rate pricing — unless you set a spend cap first. Setting a spend cap to $0 over-budget effectively pauses premium model access once the pool is exhausted. Most developers on Auto rarely need to think about this. See the full Cursor pricing page for current rates.
| Plan | Monthly price | Annual price | Usage included | Best for |
|---|---|---|---|---|
| Hobby | $0 | $0 | Limited agent requests + Tab | Evaluation and very light use |
| Pro | $20/mo | $192/yr | $20 credit pool; Auto unlimited | Professional developers at moderate daily intensity |
| Pro+ | $60/mo | $576/yr | 3× Pro on all models | Daily heavy Composer users hitting Pro limits |
| Ultra | $200/mo | Not listed | 20× Pro on all models | Multi-project background-agent power users |
| Teams Standard | $40/user/mo | $32/user/yr | Dual pools: Cursor + third-party models | Engineering teams needing admin, SSO, shared context |
| Teams Premium | $120/user/mo | $96/user/yr | 5× Standard usage | Heavy agent users on a team plan |
Pro+ ($60/month) and Ultra ($200/month): who actually needs them
Pro+ costs $60/month ($576/year) and delivers 3x the usage of Pro across all OpenAI, Claude, and Gemini models. The practical upgrade trigger is simple: if you regularly exhaust your $20 credit pool before the month ends — which happens fastest when you manually use Claude Opus or Claude Sonnet for every Composer session — Pro+ removes that friction.
Ultra costs $200/month and includes 20x the Pro usage allocation, plus priority access to new features. This tier makes sense for developers running multiple background agents simultaneously, working across several large codebases, or generating sustained high-volume code output all day. At professional consulting rates, recovering 1 to 2 hours of saved time per week covers the price difference.
One honest note: many developers who upgrade to Ultra do so because they have not switched heavy routine tasks to Auto mode. Auto mode is unlimited and handles most daily coding well. Switching from manually pinning Claude Sonnet for every request to letting Auto route — and pinning a premium model only for genuinely complex reasoning tasks — often recovers enough credits that Pro suffices. Try that before upgrading to Pro+. If you still run out, the upgrade is legitimate.
Teams pricing: the June 2026 Standard and Premium split
In June 2026, Cursor restructured the Teams plan from a single $40/user/month tier into two seats: Standard ($40/user/month, $32/user/year) and Premium ($120/user/month, $96/user/year).
Both seats include the full Teams feature set: centralized billing, SAML/OIDC single sign-on, usage behavior analytics, organization-wide privacy mode, shared team context for cloud agents, and the internal team marketplace for rules, skills, and plugins.
The June 2026 update also introduced dual usage pools for all Teams seats. Each Standard seat now includes a separate allowance for Cursor's own models (Auto and Composer 2.5) and a separate pool for third-party API models (Claude, GPT-4o, Gemini). This effectively doubled included usage for most teams at no extra cost for Standard seats — the main reason Cursor claims the changes lower costs for 90% of teams.
Premium seats ($120/user/month) include 5x the usage of Standard at 3x the price. The pitch is cost predictability: developers who would otherwise exceed Standard limits and trigger unpredictable on-demand charges can pay a known monthly rate instead. Cursor's estimate is that Premium covers a full month of intensive usage for 99% of power users.
Teams can mix seat types. A team of 10 where two developers run background agents constantly and eight use the editor for standard daily coding can buy 2 Premium + 8 Standard — $240 + $320 = $560/month — rather than $1,200/month for all Premium or $400/month for all Standard with unpredictable overages from the two heavy users.
These changes applied to new customers immediately and to existing customers on billing cycles starting July 1, 2026. For the comparison between Cursor Teams and Devin Desktop's team plans, see the Cursor vs Windsurf comparison.
How to choose the right Cursor plan
For a solo developer: run the 2-week trial seriously. Use Composer daily. Run background agents on at least one real task. If agent mode saves you meaningful time during the trial, buy Pro — $20/month is hard to argue against if it saves even 30 minutes a week at professional rates.
Pro is the right tier for professional developers coding 3 to 4 hours per day who use Auto mode for most requests. The $20 credit pool rarely runs out under Auto. If you manually use Claude Opus for every session, it will run out faster.
Pro+ makes sense only when you consistently exhaust Pro credits before month-end. Not hitting limits regularly? The $40/month difference adds up to $480/year. Don't upgrade preemptively — upgrade when the data says you should.
Ultra is for high-volume professional use: agencies running multiple active client projects, developers who live in background agents all day, and engineers whose output volume is genuinely the bottleneck. Most individual developers on a single main project do not hit Ultra-level usage.
For teams: Standard seats ($40/user) suit most engineering teams. Add Premium seats ($120/user) only for the subset of developers who run background agents heavily — typically 10 to 20% of a team, not the whole team. Cursor's Team admin panel shows per-user usage, which makes identifying who actually needs Premium seats straightforward.
For a deeper look at what Composer, Tab completion, and background agents do in practice, the Cursor AI review covers those features in full.
Cursor vs GitHub Copilot: is the price difference worth it
GitHub Copilot Pro costs $10/month. Cursor Pro costs $20/month. The $10/month gap — $120/year — is not a meaningful decision point for a professional developer. The question is whether Cursor's extra capabilities justify switching.
The capabilities that justify the price: full codebase indexing, Composer agent for multi-file tasks, and background agents for long-running work. In standardized testing from early 2026, Cursor completed a multi-component feature build in 2 rounds of prompting. GitHub Copilot Workspace needed 5 rounds with manual corrections between attempts. That gap is real on complex, multi-file tasks.
The capabilities that do not justify the price: autocomplete on straightforward, self-contained files. Copilot's completion quality is solid for that use case, and the $10/month saving is real if agent mode is not part of your workflow.
The honest framing: if you write mostly single-file scripts, maintain simple apps, or work in an IDE other than VS Code, Copilot is the better purchase. Cursor's value proposition is specifically its Composer agent and codebase context — and that value compounds as codebase complexity increases.
When not to pay for Cursor
Do not pay for Cursor if you code casually — weekend projects, occasional scripts, light automation. GitHub Copilot at $10/month covers that use case at half the price, and its quality on straightforward single-file tasks is good enough that Cursor's extra capabilities go unused.
Do not pay for Cursor if your team works across multiple IDEs. Cursor is VS Code only. Developers in JetBrains, Vim, or Visual Studio have no Cursor option — GitHub Copilot covers all of them.
Do not pay for Cursor if you are still learning to code. Cursor assumes you can read and critically review AI-generated output. Without that ability, every error the agent makes leaves you stuck with no way to trace it. Start with Lovable or Bolt to build things without needing to read the code first.
Do not pay for Cursor if your organization has strict code privacy requirements. Code is processed by third-party AI providers — Anthropic, OpenAI, Google — and there is no self-hosted option. Cursor's Privacy Mode prevents code storage but does not prevent code from leaving your machine entirely. Verify this is acceptable with your security team before rolling out organization-wide.
Ready to try Cursor?
Free plan (limited). Pro at $20/month.
Frequently Asked Questions
Continue reading
July 19, 2026