Cursor is an AI-native code editor that routes your queries through frontier models including Claude, GPT-4, and Gemini. Pricing changed significantly in mid-2025 when Cursor moved from a request-based model to a credit-based system. Understanding how credits deplete is more important than knowing the headline plan price.
What are the Cursor AI pricing plans?
| Plan | Monthly Cost | Monthly Credits | Best For |
|---|---|---|---|
| Hobby | Free | None | Trying Cursor, light users |
| Pro | $20 ($16 annual) | $20 | Most professional developers |
| Pro+ | $60 | $70 | Daily agent users |
| Ultra | $200 | $400 | Agent power users |
| Teams | $40/user | Pro-equivalent per seat | Engineering teams of 3+ |
Annual billing saves approximately 20% across all paid tiers. A 7-day free Pro trial is available without a credit card.
How does the Cursor credit system work?
Cursor runs two separate credit pools that reset monthly. Understanding which pool a given action draws from is the key to reading your bill accurately.
| Pool | What draws from it | Rate | Monthly included |
|---|---|---|---|
| Auto + Composer | Auto mode, Composer 2.5 | $1.25/M input · $6.00/M output · $0.25/M cache read | Unlimited on paid plans |
| API | Manual model selection, Max mode, Premium routing | Model's published API rate (e.g. Claude Sonnet 4: $3/M in · $15/M out) | Equal to plan price: $20 Pro · $70 Pro+ · $400 Ultra |
Auto mode routes each request to a cost-efficient model and draws from the cheaper Auto+Composer pool at flat rates — not from your monthly API credit budget. This is what makes it effectively unlimited for practical purposes: individual requests are inexpensive and the pool is not capped.
Composer 2.5 also draws from the Auto+Composer pool, not the API pool — so it costs significantly less than manually selecting an equivalent frontier model. This surprises many users who assume all non-Auto selections hit the same budget.
Manual model selection — choosing Claude Sonnet, GPT-4, or any specific model from the dropdown — draws from your monthly API credit pool at that model's API rate. Claude Sonnet depletes this pool roughly twice as fast as Gemini under equivalent workloads.
Max mode extends the context window to the model's maximum and draws from the API pool at token-based pricing. A handful of Max mode sessions on a large codebase can exhaust a Pro plan's $20 API credit budget in a single day.
Tab completion does not draw from either pool on paid plans.
If you exhaust your monthly API credit pool, Cursor offers pay-as-you-go overage billing at the same API rates with no markup, or you can upgrade to a higher-tier plan.
How much does Cursor cost for different types of developers?
Casual users who rely on Tab completion and occasional Chat will rarely exhaust a Pro plan's credits. Many casual users stay within the Hobby free tier.
Daily professional developers using Chat and Agent mode regularly will find Pro ($20/month) sufficient if they stick to Auto mode for most tasks. Developers who manually select premium models for agent-heavy work typically need Pro+ ($60/month).
Power users running Cursor agents as a primary development workflow will find Pro+ or Ultra most cost-effective. Ultra's $400 credit pool covers heavy multi-agent use without the overhead of calculating API overages.
Teams get Pro-equivalent AI access per seat at $40/user/month, with centralized billing and shared context across the organization.
What are common mistakes to avoid?
- Subscribing to Ultra immediately before determining if Pro is sufficient
- Manually selecting the most capable model for every task regardless of complexity
- Using Agent mode for tasks that Chat handles adequately
- Bloated
.cursorrulesfiles that add thousands of tokens to every request - Not enabling annual billing (saves 20% on any paid tier)
