Why did Cursor grade D on churn?

Cursor scored 42 out of 100. The June 2025 pricing restructure is the defining trust event: Pro users who were paying $20/month went from 500 effective requests to around 225 overnight. The CEO publicly apologized and issued refunds, but migration to Claude Code and Windsurf had already accelerated. The grade reflects the trust damage, not the product quality.

What happened with Cursor's June 2025 pricing change?

Cursor introduced a credit system that effectively cut Pro tier usage in half at the same monthly price. Users reported going from 500 requests to roughly 225 at $20/month. Public outcry on Hacker News, Reddit, and X led to the CEO issuing a public apology and offering refunds. The credit-system framing survived the apology.

Is Cursor worth it in 2026?

It depends. The core editing experience still ranks well in public benchmarks. The trust problem is pricing volatility and credit-system opacity. Users who value predictable cost per AI request have migrated to Claude Code (with API-based pricing) or Windsurf. Users who prioritize editor integration and do not mind the credit system still use Cursor.

What are Cursor alternatives after the pricing change?

The two most-cited in migration commentary: Claude Code (CLI, API-priced, no credit system), Windsurf (IDE-based, formerly Codeium). GitHub Copilot saw return-migrations from Cursor users specifically because Copilot pricing did not change. Zed also comes up for Rust and Go developers.

Why was a $9B company so fragile on pricing?

The $9B valuation was recent and investor-driven, not retention-proof. SaaS fundraising at late-stage scale creates pressure to increase ARR per user. Cursor's pricing restructure in June 2025 followed the pattern of monetizing existing customers rather than acquiring new ones. The backlash was predictable in hindsight. The speed of the migration to alternatives was not.

Cursor's $9B Pricing Mistake: Churn Teardown

A RetentionCheck churn teardown of Cursor

TL;DR

Grade

D (42 / 100)

Sample

40+ public complaints (HN, Reddit, X)

Top driver

June 2025 pricing restructure (credit system)

CEO apologized, refunds issued, and migration to Claude Code and Windsurf accelerated. Valuation peaked at $9B just before trust collapse.

Methodology: 40+ public complaints from TechCrunch coverage, two Hacker News threads (Cursor IDE support hallucination and Cursor 1.0), Reddit discussion of the June 2025 pricing change aggregated through pricing analyses, InfoQ coverage of Cursor 3 community reaction, and eesel AI's Anysphere review. Analyzed with RetentionCheck.

Cursor is a $9 billion AI-coding tool that made its most loyal users feel deceived in a single pricing decision. The Churn Health Score reflects it.

The Churn Health Score

Cursor scored 42/100, grade D. One critical driver, two high, two medium, one low.

The critical driver is specific and dated: June 16, 2025, the day Cursor shifted Pro pricing from 500 fast responses per month to "$20 of usage billed at API rates." Effective capacity dropped roughly 50% for heavy users. The CEO apologized publicly. Refunds were issued. The backlash was documented by TechCrunch.

The 5 churn patterns

1. June 2025 pricing restructure (critical, 90% confidence)

On June 16, 2025, Cursor changed the $20/month Pro plan. What was 500 fast responses plus unlimited slow responses became $20 worth of API-rate usage. Effective requests for heavy workflows dropped from approximately 500 to approximately 225 per month. Reddit erupted. Twitter picked it up. TechCrunch covered it on July 7.

"I went back to Copilot after the June credit change. Cursor Pro went from 500 to about 225 effective requests at $20 and Windsurf gives me comparable context at $15." — Reddit, aggregated

One Hacker News commenter summarized the industry concern: with $900 million in funding, the company should be able to absorb compute costs rather than pass them through to users. That framing stuck.

Cursor issued refunds. The CEO apologized. The trust damage compounded anyway.

2. AI support hallucinated a lockout policy (high, 85%)

In April 2025, Cursor's AI customer support fabricated a non-existent policy telling users they were blocked from multi-device login. Users canceled based on the false information before Cursor could clarify. A Hacker News thread titled "Cursor IDE support hallucinates lockout policy, causes user cancellations" became a case study in AI-without-humans support backfiring at the trust level.

The lesson cuts beyond Cursor. Any AI-support system that confidently fabricates policy will damage trust faster than a slow human response would.

3. Opaque credit counter anxiety (high, 80%)

Post-restructure, users consistently describe the credit counter as anxiety-inducing and opaque. Cannot predict when they will hit the limit. Usage tracking lags behind actual spend. Developers who signed up expecting flat-rate monthly pricing found themselves monitoring a meter.

"The credit counter is anxiety-inducing and opaque." — Reddit

Flat-rate pricing is a feature. Unpredictable usage metering is a cost.

4. Cursor 3 agent-first redesign polarized the community (medium, 68%)

April 2026 Cursor 3 introduced an agent-first interface, explicitly moving beyond the traditional IDE model. Per InfoQ coverage, community reaction divided. Developers who chose Cursor specifically for the Cursor-plus-VS-Code experience are re-evaluating whether the new direction still serves their workflow.

Bold redesigns are sometimes necessary. They are also sometimes losing bets. Keeping the classic IDE-first mode available alongside agent mode would let users choose instead of forcing the change.

5. Claude Code and Windsurf gaining migration velocity (medium, 72%)

Post-pricing-change, alternatives absorbed measurable migration. Windsurf at $15/month was cited repeatedly as the easiest transition. Claude Code (from Anthropic) attracted heavy-spend users with API-native pricing.

"$2k a week with premium models on Cursor. Switched to Claude Code Max. Equally as prolific and paying 1/10th the price." — Reddit commenter who migrated

Cursor's moat was always depth of IDE integration, not lowest per-token price. Competing on price against API-native tools is a losing game.

What Cursor does right

HN and Reddit commentary consistently praises Cursor's IDE depth, codebase-wide context, and integration polish. Developers who stay cite the raw productivity boost of multi-file context-aware AI. The product at its best is regarded as category-defining. The $9B+ valuation reflects real product-market fit on the core use case.

The churn story is not "Cursor is a bad tool." It is "Cursor's monetization strategy and trust hygiene are eroding the loyalty of the customers who made it famous."

3 things Cursor could fix

Publish a 12-month pricing stability commitment for Pro. Grandfather existing customers on prior credit allowances. Separate power-user tier with pay-as-you-go so base Pro stays predictable.
Human review gate on AI-generated policy claims. AI support can answer questions. Humans confirm anything involving policy, billing, or account access.
Real-time credit display with next-action cost estimates. Users should never be surprised by their bill. Soft-limit warnings before cutoff. Monthly cap settings users control.

What this means for your SaaS

Cursor's story shows how a single pricing decision can compound into a multi-month trust tax. The June 2025 restructure was not unreasonable on the merits (AI costs genuinely rose). The communication failed. The grandfathering failed. The credit system failed. And the alternatives were close enough that migration became viable.

Every SaaS pricing change is a trust transaction. The revenue gain compounds against the permanent trust loss. Cursor is now the case study.

If you have cancellation feedback sitting in a spreadsheet, run it through RetentionCheck in 30 seconds. The pattern is usually underneath the individual reasons.

Key takeaways

A single pricing restructure can define a company's churn story for a year or more.
Flat-rate pricing is a feature. Unpredictable usage metering is a cost.
AI customer support that fabricates policy is a trust grenade. Human review gates exist for a reason.
Well-funded companies absorbing compute costs instead of passing them through keeps the developer community loyal.
Competing on price against API-native tools (Claude Code) is a losing game. Compete on depth of integration.

Sources

Related teardowns

Asana + Notion: Both Graded F (Grade F) · 15 Trustpilot reviews, billing trust collapse
Notion's Mid-Life Crisis (Grade D) · 60+ HN complaints, feature bloat
Why Teams Are Leaving Linear (Grade B) · Narrow, bounded churn (the opposite story)
Figma's 2025 Pricing Hike (Grade C) · +33% + Penpot surge

Brian Farello is the founder of RetentionCheck, an AI-powered churn analysis tool for SaaS teams. Try it free at retentioncheck.com/try. Other teardowns: Notion (D), Linear (B), Figma (C), Asana (D), Monday.com (C).