Cursor (Anysphere)

About

Cursor is an AI-native code editor built by Anysphere, founded in 2022 by four MIT students (Michael Truell, Sualeh Asif, Arvid Lunnemark, Aman Sanger). The product launched publicly in March 2023 as a fork of Visual Studio Code with AI features baked in at the editor level rather than bolted on as an extension.

By early 2026 the company had crossed $2 billion in ARR, reached a $29.3 billion valuation (November 2025 round led by a16z, Thrive Capital, Benchmark), and grown to over 300 employees — making Cursor one of the fastest-revenue-ramping software companies on record.

The product competes with GitHub Copilot, Codeium, Continue, and Cline, but its pricing model is structurally different: where Copilot bundles AI into a flat seat fee, Cursor exposes underlying API costs to the user and takes margin on routing.

Pricing summary : How Cursor’s hybrid credit-pool model works

Cursor uses a hybrid model with two dimensions:

1. Platform access (seat-based): $20–$200 per user/month depending on plan, or $40/user/month for Teams.
2. AI inference (credit-based, $1 credit = $1 of model API cost): Each paid plan includes a monthly credit pool equal in dollars to the plan price. Premium model requests draw from the pool. Auto mode is unlimited.

What makes this different: Most AI coding tools (Copilot, Codeium) charge a flat seat fee and absorb all model costs. Cursor exposes the underlying token economics — users see exactly how much each request costs and can choose between flat-rate Auto mode and per-token premium models.

See the six-plan grid at the top of this page for the full lineup. Plans split into individual (Hobby, Pro, Pro+, Ultra) and business (Teams, Enterprise) ladders.

Pricing by product

Editor (Individual plans)

Editor (Business plans)

Bugbot (separate product)

Sales motions across products: PLG / self-serve for Hobby through Teams; sales-led for Enterprise.

Per-model pricing inside the credit pool

When you select a specific frontier model (not Auto), credits deplete at posted per-million-token rates:

Cache hits cost 10% of standard input price. Max Mode (up to 1M token context) carries roughly a 20% premium on the same rates.

A $20 credit pool gets you roughly 500 GPT-5.1 Codex requests or 45 Claude 4 Opus requests under typical workload assumptions — a ~10x spread between cheapest and most expensive frontier model.

Hidden costs : What heavy individual and team users actually pay

The advertised $20/mo headline understates what heavy users actually pay. Two real-world examples:

Heavy individual developer (Claude Opus + multi-file refactors)

A developer hitting Claude Opus daily for refactors and architecture pays roughly 5x the headline. The Ultra plan ($200 with 20x credits) becomes economically rational once monthly spend crosses ~$150 — but until you’ve spent that 5x premium for a few months, you don’t know you needed Ultra.

10-person Teams workspace

For teams, the per-user variance is the budgeting challenge: one engineer running long-context Opus sessions can spike a team bill by 30% in a week. See our usage-based pricing thresholds and alerting guide for how to architect spend governance for variable-cost AI tools.

Want to estimate your own Cursor bill? Use the Cursor pricing calculator to model your monthly cost based on plan, model mix, and request volume.

Pricing evolution : From generous Pro to the June 2025 credit reset

Cadence

Tracked range: 2024 Q2 – 2026 Q2. Quarters not listed above were verified stable (0 price changes, 0 SKU additions).

Notable changes

• 2024 Q4 — Page wording shifts from “fast requests” to “premium requests”; o1-mini surfaces as a priced-model option. Plan prices unchanged.
• 2025-05-13 — Blog post: “Small improvements to pricing”. Usage-based models become payable from included quota; MAX long-context option moves from per-tool-call to API pricing. Framed as a transparency improvement.
• 2025-06-16 — Pro plan changes from “500 requests” to “$20 of model inference at API prices.” Ultra $200/mo introduced simultaneously (20× Pro credit pool). Existing users could opt to stay on the legacy 500-request method.
• 2025-07-04 — CEO Michael Truell publicly apologizes for the rollout’s unclear communication. Refunds offered for unexpected charges between mid-June and early July. Sets new policy: pricing changes flagged 30+ days in advance.
• 2025-08 — Teams pricing dropped fixed per-request costs and unified with individual-plan variable billing. Bugbot launched as separate $40/user/mo product.
• 2025-09 — Pro+ $60/mo introduced between Pro and Ultra (3× usage), filling the gap surfaced by the June/July incident reports.
• 2026-02 — Lineup formalised at Hobby / Pro / Pro+ / Ultra / Teams / Enterprise. Annual billing introduced at 20% discount.

The June 2025 incident in detail

On June 16, 2025, Cursor switched from request-based to credit-based billing without a clear advance email to users. The new model meant:

• A multi-file refactor using Claude Opus could cost $5–$15 per session in credits, where previously it counted as a single “fast request.”
• Several high-profile users posted receipts showing daily charges of $10–$20 they hadn’t anticipated.
• One widely-shared thread documented a team whose $7,000 annual subscription was depleted in a single day of normal usage.

The July 4, 2025 apology committed to: (1) refunds for unexpected charges, (2) better in-product spending visibility, (3) email warnings before pricing changes. This response — the public apology + refund — became a reference point cited by Anthropic, OpenAI, and others when communicating their own pricing changes through 2025–2026.

The June 2025 Incident in Detail

On June 16, 2025, Cursor switched from request-based to credit-based billing without a clear advance email to users. The new model meant:

• A multi-file refactor using Claude Opus could cost $5–$15 per session in credits, where previously it counted as a single “fast request.”
• Several high-profile users posted receipts showing daily charges of $10–$20 they hadn’t anticipated.
• One widely-shared thread documented a team whose $7,000 annual subscription was depleted in a single day of normal usage.

The July 4, 2025 apology committed to: (1) refunds for unexpected charges, (2) better in-product spending visibility, (3) email warnings before pricing changes. This response — the public apology + refund — became a reference point cited by Anthropic, OpenAI, and others when communicating their own pricing changes through 2025–2026.

What’s unique : Auto mode, transparent pass-through, six-plan stratification

1. Transparent pass-through. $1 of credits = $1 of underlying API cost. Most competitors bundle inference into seat fees; Cursor exposes it. This makes it the only consumer-facing AI tool where you can directly compare “what GPT-5 actually costs me” against the API.

2. Auto mode as a category innovation. A flat-rate router that absorbs model selection — users get unlimited usage and Cursor takes margin on routing arbitrage. No other major AI tool offers this primitive.

3. Credit pool denominated in dollars. Cursor’s “$20 includes $20 of credits” is unusually legible. Compare to Vercel ($20 credit but multi-dimensional metering across bandwidth, requests, CPU, memory) or Intercom ($0.99/resolution with no included quantity).

4. Six plans for one product surface. Most AI tools ship 3 tiers (free / pro / enterprise). Cursor ships six — the granularity reflects how steeply demand stratifies in AI coding ($20 casual users vs $200 power users vs custom enterprise).

Strengths & weaknesses

Billing UX : Dashboards, hard caps, and post-incident alerts

• Spending dashboard — shows real-time credit consumption per model, per day.
• Hard cap — users can set a maximum monthly spend; once hit, premium model requests fail (Auto mode continues).
• Email alerts at 50% / 75% / 100% of credit pool consumption (added post-July 2025 apology).
• No spend by default beyond plan — overages require explicit opt-in since the August 2025 update.
• Per-user spending visibility on Teams — admins see consumption breakdown by member.

Strategic wins : Why the credit pool and Auto mode landed despite the rollout

1. Auto mode aligns vendor and customer incentives

The flat-rate router takes margin only when Cursor’s routing decisions are good. Bad routing → Cursor eats the cost. This is the rare case where vendor profit and customer experience point the same direction. Most outcome-based pricing requires the vendor to define an “outcome”; Auto mode lets the underlying API cost define it.

2. The credit-pool model is honest

$1 credit = $1 underlying API cost is the most transparent unit economics in the AI tooling category. Compare against competitors that bundle inference into seat fees with implicit 2-5x markups (the token markup pattern is endemic). Cursor’s transparency is a defensible differentiator once trust is rebuilt.

3. Multi-year frontier-model deals fund the Ultra promise

The November 2025 Ultra launch wasn’t possible without negotiated commitments from OpenAI, Anthropic, Google, and xAI for predictable volume pricing. Competitors without those deals can’t offer “predictable price for very heavy usage.”

4. Refund-then-fix posture beat radio silence

The July 4 2025 apology + refund pattern is now the playbook for pricing change communication. The cost was real but bounded; the long-term trust dividend is significant.

Areas to improve : Annual commits, spend projections, Bugbot positioning

1. Lead with TCO scenarios, not plan cards

The pricing page shows the six plans without surfacing the real bill for archetypal users. A “heavy individual” archetype showing “$20 plan + $80 typical overage = $100/month” would set expectations honestly. Today the headline-vs-reality gap creates bill-shock churn.

2. Offer annual commit discounts on Pro/Pro+/Ultra

Finance-led buyers used to seeing 10–20% off-list for annual commitment have to choose between predictability and discount. Cursor offers neither. A simple “20% off when paid annually with rollover credits” would unlock the long-tail of teams currently sitting on legacy Pro.

3. Ship per-tier spending forecasts

The spending dashboard shows what you’ve spent. It doesn’t project what you’ll spend at current pace. A “your trajectory: $X by month-end” projection would let users course-correct before hitting the cap. See our guide to spend caps and alerting for what good predictability looks like.

4. Define the “Bugbot economics” publicly

Bugbot at $40/user/mo flat is unusual — typically per-user team pricing is the same or slightly less than individual. The flat $40 across team and individual suggests either Cursor wants to nudge individuals toward Teams for shared admin, or that Bugbot’s costs are dominated by per-PR inference rather than headcount. Make this explicit.

5. Standalone Bugbot product surface

Bugbot ships as a separate product but lacks a dedicated pricing page. Mirroring Intercom’s standalone Fin strategy, publishing a stand-alone Bugbot page that integrates with non-Cursor IDEs would unlock the long tail of teams who don’t want to switch editors.

Key takeaways

1. Transparency requires communication. $1 credit = $1 API cost is honest, but Cursor’s June 2025 rollout proved that legibility without 30+ day announcement is functionally opaque.
2. A flat-rate option absorbs unit-economic risk. Auto mode is what makes the variable credit pool palatable. Without it, the same pricing would feel hostile.
3. Refund > defensiveness when you mispriced. Cursor lost ~$15M in refunds; competitors who tried to ride out similar episodes lost orders of magnitude more in churn.
4. Six plans is not too many if user volume genuinely stratifies. The $20 / $60 / $200 / $400 spread reflects real consumption variance among professional developers, not synthetic upsell pressure.
5. Charge for outcomes when possible, transparent inputs when not. Cursor’s mix is “Auto mode = outcome (working completion), per-model = transparent input.” A reasonable design when both modes coexist.

UBP implications

1. Expose unit economics carefully. Cursor’s credit-to-dollar mapping is legible, but the migration from “fast requests” to “credits” required users to relearn what their workflows cost. Transparency without communication isn’t transparency.
2. Reserve a flat-rate option for predictability-seekers. Auto mode is the safety valve for users who don’t want to think about model choice. Without it, the credit pool would feel like an air-conditioner running with the meter on.
3. Communicate pricing changes 30+ days in advance. The Cursor incident cost real trust. Every major AI tool that changed pricing in late 2025 cited Cursor’s playbook for what NOT to do.
4. Bill in the unit your customer thinks in. Developers think in “requests” and “completions,” not tokens. Cursor’s per-token rates required user education and produced sticker shock even when total costs were lower.

Sources

• Cursor official pricing page (accessed 2026-05-23)
• Cursor docs: Models & Pricing (accessed 2026-05-23)
• Cursor blog (accessed 2026-05-23)
• Cursor changelog (accessed 2026-05-23)

Bottom line

Cursor proved that variable pricing can work for developer tools — but only if the unit is something the user can actually reason about. The credit pool succeeded as a model. The June 2025 rollout failed as a communication. Every AI tool that follows now ships with 30-day notice, in-product spending visibility, and an Auto-mode safety valve — and that’s largely Cursor’s legacy.

Want to see how other AI coding tools compare? Browse the pricing blueprint for the full corpus.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Six-Plan Structure Formalized

Feb 2026

Lineup stabilized at Hobby (free), Pro ($20), Pro+ ($60), Ultra ($200), Teams ($40/user), Enterprise (custom). Annual billing introduced at 20% discount.

Ultra Plan Introduced

Nov 2025

$200/mo Ultra tier launched offering 20x Pro's credit pool plus priority access to new features. Positioned for heavy agent users.

Teams Moves to Variable Billing

Aug 2025

Teams pricing dropped fixed per-request costs and moved to the same variable API-based billing as individual plans. Auto mode included usage was adjusted.

Public Apology + Refunds

Jul 2025

CEO Michael Truell issued a public apology for the unclear rollout. Cursor offered refunds for unexpected charges incurred between mid-June and early July 2025.

Switch to Credit Pool Pricing

Jun 2025

Pro plan $20/mo now includes a $20 credit pool tied to underlying API costs. Different models consume credits at different rates. Users could exceed the pool and incur overages. Rollout was rocky — many users hit unexpected daily charges of $10–$20.

Pro Plan at $20/month (Request-Based)

Jan 2024

Generous tier with monthly allocation of 'fast' premium model requests plus unlimited 'slow' fallback. Considered one of the most generous in the AI tooling category.

Cursor Launch

Mar 2023

Anysphere launches Cursor as a VS Code fork with inline command and chat. Initially free during private beta.