Updated July 2026

Cursor Pricing 2026: Plans & Real Token Costs

Five tiers from $0 to $200. Each paid plan bundles a credit pool equal to its price, and Auto mode is unlimited. Here's the complete breakdown — and how to cut your bill.

$20/moPro plan
$20 poolincluded model credit
UnlimitedAuto mode included
40–70%prompt cut with Terse

Cursor Plan Tiers — July 2026

Cursor has five plans in 2026. Every paid tier includes a monthly usage credit pool equal to the plan price, plus unlimited Auto-mode model usage that never touches that pool.

Plan Price Credit Pool Auto Mode What's Included
Hobby $0 Limited Basic Tab & Agent limits, try it free
Pro $20/mo $20 included Unlimited Unlimited Tab, extended Agent, Cloud Agents, all frontier models, MCP, hooks
Pro+ $60/mo $60 included Unlimited Everything in Pro, 3× the frontier-model credit pool
Ultra $200/mo $200 included Unlimited Everything in Pro+, 10× the credit pool for heavy frontier use
Teams $40/user/mo Per-seat pools Unlimited Pro features, admin & billing, org privacy, split usage pools

Auto mode — where Cursor picks a cost-efficient model — is included at no extra cost. Manually selecting frontier models (Claude, GPT, Gemini) draws from your credit pool; overages bill at API rates with no markup.

Real Token Rates Under the Hood

Cursor bills by tokens beneath the plan pool. These approximate Auto-mode rates are what your credit pool draws down against, and what overages are charged at.

Usage Type Approx Rate ($/M tokens) Counts Against Pool? Notes
Auto mode (Cursor picks model) Included No Unlimited on all paid plans
Cache read ~$0.25 When frontier-selected Cheapest — reused context
Input tokens ~$1.25 When frontier-selected Prompts, files, context
Output tokens ~$6.00 When frontier-selected Model responses — most expensive

Because Auto usage is included, these rates mainly matter when you manually run Claude, GPT, or Gemini past your credit pool and start paying overages.

Real Monthly Cost by Usage Pattern

What each kind of developer actually spends. The plan price is only a floor — heavy frontier-model use pushes you into overages on top.

Light / Auto-First
$20
per month
Lives in Auto mode (included)
Rarely hand-picks frontier models
$20 Pro pool never fully spent
Effectively flat $20/month
Typical Developer
$20–$60
per month
Auto for routine, Claude/GPT for hard tasks
Drains the $20 pool most months
Modest overages or upgrades to Pro+
Multi-file refactors, agent runs
Frontier Power User
$100–$300+
per month
Hand-picks Opus/GPT on large contexts
Blows through the pool by mid-month
Needs Ultra ($200) or heavy overages
Long agent runs, big repos

Where Your Credit Pool Goes

A single day of hand-picked frontier-model agent work can drain the entire $20 Pro pool. Here's a realistic day when you keep selecting Claude Opus manually on a large context — and how much smaller it gets with tighter prompts.

Frontier mode, verbose — one heavy day
Large context re-sent each turn (input)~$8.00
Duplicate file / folder drags (input)~$4.50
Long agent output (output tokens)~$9.00
Pool spent~$21.50
Blows past the whole $20 Pro pool in a day → overages
With Terse + Auto discipline — same work
Compressed prompts (40–70% shorter input)~$3.00
Trimmed context, no duplicate drags~$1.20
Routine turns on Auto (included, $0)~$0.00
Pool spent~$4.20
$20 pool lasts the whole week instead of one day

How to Reduce Your Cursor Bill

Six proven techniques — starting with the ones that have the highest impact per hour of effort.

🗜️

Compress Prompts with Terse

Terse's 7-stage pipeline removes filler, fixes typos, and strips redundancy from the prompts you type into Cursor before they hit the model. 40–70% fewer input tokens means your credit pool drains far slower on frontier models.

Highest Impact
🔀

Default to Auto Mode

Auto mode is unlimited and included on every paid plan — it never touches your credit pool. Let Cursor route routine edits, searches, and boilerplate to a cost-efficient model, and reserve your pool for tasks that truly need Claude or GPT.

Included, $0
🎯

Reserve Frontier Models

Manually selecting Opus-class models is where bills spike. Hand-pick them only for genuinely hard reasoning or architecture work. A single day of frontier agent runs on a big repo can consume the entire $20 Pro pool.

Avoid Spikes
📎

Trim Your Context

Dragging whole folders into chat re-sends thousands of input tokens every turn. Attach only the files you need. Smaller context means cheaper cache reads and less pool burn — especially on frontier models billed per token.

Cuts Input Tokens
🔍

Track Per-Request Cost

Terse's Agent Monitor surfaces per-turn token usage and cost in real time, so you can see which requests are draining your credit pool and switch back to Auto before an expensive turn repeats.

Catch Expensive Turns
📐

Right-Size Your Plan

If you consistently exhaust the $20 Pro pool mid-month, Pro+ ($60) or Ultra ($200) may be cheaper than paying overages. If you rarely leave Auto, Pro is plenty. Match the plan to how often you hand-pick frontier models.

Avoid Overages

Frequently Asked Questions

Answers to the most common questions about Cursor costs in 2026.

How much does Cursor cost?
Five tiers: Hobby $0, Pro $20/mo, Pro+ $60/mo, Ultra $200/mo, and Teams $40/user/mo. Every paid plan includes a monthly credit pool equal to its price (Pro = $20), and Auto mode is unlimited and included even after the pool is spent.
What is the credit pool and how does it work?
Each paid plan bundles a credit pool equal to the plan price. Auto mode usage is free and doesn't draw from it. When you manually pick a frontier model (Claude, GPT, Gemini), that usage is billed by tokens against the pool. Empty the pool and you either switch to Auto or pay overages at API rates with no markup.
Why did my Cursor bill spike?
Almost always from manually selecting frontier models on large contexts. Token billing means big repos, long agent runs, and duplicate file reads burn the pool fast — a full day of hand-picked Opus agent work can blow through the entire $20 Pro pool and push you into overages. Auto mode never causes this.
Is Cursor worth it vs paying API rates directly?
For most developers, yes. Overages bill at API rates with no markup, and Auto mode is unlimited — so Pro's $20 buys unlimited Auto plus $20 of frontier credit plus the editor. Only if you hand-pick Opus-class models on huge contexts all day would a bare editor on direct API billing be cheaper, and you'd lose Tab, Cloud Agents, and routing.
How do the Teams pools work in 2026?
Teams is $40 per user per month. In 2026, each seat's usage is split into two pools: one for Cursor's first-party models (Composer/Auto) and one for third-party API models (Claude/GPT/Gemini). This lets admins see and manage frontier-model spend separately from included Auto usage.
How do I reduce my Cursor costs?
Default to Auto (included, no pool draw), reserve frontier models for hard problems, and use Terse to compress the prompts you type into Cursor by 40–70% before they consume input tokens. Trim context so you don't re-send whole folders, and use Terse's Agent Monitor to catch expensive turns before the pool drains.

Also Works With

Terse reduces token costs across every major AI coding tool.

🧠 Claude Code — Terminal Agent 💬 ChatGPT — Browser 🤖 GitHub Copilot — VS Code 🖥️ Aider — Terminal Agent

Make Your Cursor Credit Pool Last 3× Longer

Terse compresses prompts, trims context, and monitors per-request cost so your $20 pool stretches across the week instead of one heavy day. 30-day free trial — no credit card required until your trial ends.

Start Free Trial — macOS Calculate Your Savings