CHICustomer Hostility Index
← Back to Companies

Cursor · CHI Company Investigation · AI

CursorAnysphere, Inc. · AI code editor · Subscription plans

Conclusion: Partially supports · Mechanism: a fixed allowance made variable, then unpublished · No score assigned

Cursor kept Pro at $20. What changed was how much $20 promised.

The price stayed fixed. The unit of value did not. Cursor moved Pro from a fixed request allowance with an unlimited fallback to usage-based metering, applied the June 2025 switch to existing subscribers by default, and later apologised that the change was not communicated clearly. Today its plan pages no longer publish how much usage $20 includes.

$20Pro per month on every archived page, May 2023 – Oct 2026
500Requests a month on Pro until June 2025
16 Jun 2025Existing Pro users moved by default
UnpublishedIncluded usage on current plans
● Estimated read 10 min
01 · What $20 bought

Five versions of the same $20 plan.

Each step below is dated from Cursor’s own pages, archived copies of them, or posts by Cursor staff. The price line never moves. “Promised” here means what Cursor’s published plan wording said $20 included, not what any individual customer received.

  1. Before 16 Jun 2025 $20 / month 500 requests + unlimited slow

    “500 requests per month” and “Unlimited slow requests”. Paying for more was opt-in.418

    Published caveatSlow requests “may occasionally be throttled during periods of high demand.”4

  2. 16 Jun 2025 $20 / month “Unlimited-with-rate-limits”

    Cursor said it was making Pro “more generous”. Existing users were moved by default, with a temporary option to switch back.78

    Not in the postAny dollar allowance, and any mention of Auto.7

  3. 30 Jun – early Jul 2025 $20 / month At least $20 of usage + unlimited Auto

    “at least $20 of model inference at API prices per month”, plus “unlimited access to the ‘Auto’ model”.10

    Date30 June is the date Cursor gives for the edit. Archived copies show it between 1 and 3 July.1011

  4. Renewal after 15 Sep 2025 $20 / month Auto now draws on the $20

    Included usage remained. Auto began to consume it at published token rates, from each subscriber’s first renewal after 15 September.1617

    Annual plansKept unlimited Auto until their annual renewal.16

  5. From late Aug 2026 $20 / month “Included”

    The dollar figure left Cursor’s documentation. Pro is described as “Extended limits on Agent”.232426

    Still onlineA February 2026 Cursor blog post still says “at least $20”.28

Earlier history. Pro was already $20 a month on the earliest archived pricing page opened, 13 May 2023, when it was a token-based plan.1 By July 2023 it read “500 fast GPT-4 uses per month” and “Unlimited slow GPT-4 uses”.2

Cursor kept Pro at $20. What changed was how much “$20” promised.

A fixed count with an unlimited fallback became metered usage, then metered usage with the fallback metered too, then an allowance whose size is no longer stated. That did not leave every customer with less: on Cursor’s own figures, users of cheaper models could get more usage than the old 500-request count (section 04).

02 · The June 2025 migration

The change arrived before the explanation did.

This is the strongest adverse event in the record. Cursor announced the new Pro model and applied it in the same week, to existing subscribers, under wording it later said was unclear.

  1. 11 Jun
    Before: $20 for 500 requests

    Pricing page: “500 requests per month”, “Unlimited slow requests”.4

  2. 16–17 Jun
    Announced and applied the same week

    “We’re also happy to report we’re rolling out changes this week to make our Pro plan more generous. By default, the Pro plan will now follow an unlimited-with-rate-limits model, and all limits on tool calls will be lifted. Existing users can choose to stay with the “500 request limit” method if they prefer.”7 The pricing page changed between 16 June 17:32 and 17 June 07:42 UTC.56

  3. 18 Jun
    Existing users moved by default

    Cursor’s docs: “Standard Pro Plan users will be automatically moved to this new Pro Plan”. A staff member later confirmed “all users were moved over automatically”. An opt-back sat in the dashboard settings.89

  4. 30 Jun
    The post is rewritten

    “Note: We updated this blog on June 30 to improve its clarity.” The new sentence named “at least $20 of model inference at API prices per month” and “unlimited access to the ‘Auto’ model”.10

  5. 4 Jul
    Apology and refunds

    Cursor wrote that the changes “were not communicated clearly”, took responsibility, and offered refunds for unexpected usage between 16 June and 4 July.12

  6. 8–14 Jul
    Opt-back button removed

    Reported gone on 8 July; a staff member confirmed its removal on 14 July and pointed users to email. Staff said on 10 July the option “was stated to only be available for a limited time”. We found no earlier Cursor page stating a time limit.1314

16 June 2025 · the announcement
“By default, the Pro plan will now follow an unlimited-with-rate-limits model”
“Introducing the Ultra Plan”, archived 17 Jun 20257
4 July 2025 · the clarification
“We were not clear that “unlimited usage” was only for Auto and not all other models, which have at least $20 of included usage.”
“Clarifying Our Pricing”, archived 5 Jul 202512

“Our recent pricing changes for individual plans were not communicated clearly, and we take full responsibility.”Cursor, “Clarifying Our Pricing”, 4 July 202512

Notice. We found no advance notice of the 16 June change in the archived pages we opened. Whether subscribers were emailed could not be verified either way. How many customers had unexpected charges, and how many were refunded, is not public. Refunds were requested by email: “Please contact pro-pricing@cursor.com if you had a surprise usage bill”.12

03 · The free fallback

Cursor replaced one unlimited fallback with another, then later metered the replacement.

What happened after a Pro subscriber used up the included allowance is the clearest case of something once included becoming metered.

Phase 1 · to June 2025

Unlimited slow requests

  • On the Pro plan, with no extra charge4
  • Could be throttled; Cursor’s page said so4
  • Did not cover every frontier model: Claude Sonnet 4 was “Currently not available in slow pool”1820
  • Ended for users moved to the new plan
Phase 2 · from the 30 June edit

Unlimited Auto

  • First stated in the edit Cursor dates to 30 June 202510
  • Cursor’s model router, offered without limit once the allowance ran out12
  • “Since June 2025, Auto has been unlimited for individuals”, Cursor later wrote16
Phase 3 · renewal after 15 Sep 2025

Auto becomes metered

  • Announced 12 August 202516
  • “At your next billing renewal after September 15, Auto will contribute to your included monthly usage”16
  • At $1.25 input, $6.00 output, $0.25 cached input per million tokens17
  • Annual subscribers kept unlimited Auto until annual renewal16
From a free fallback, to a different free fallback, to metered usage in about three months.

On 22 July 2025, three weeks before the announcement, a Cursor staff member wrote on its forum: “Auto remains free of charge for users on all paid plans!”15 Cursor’s own posts described Auto as “unlimited”, not “free”. The August change gave about five weeks’ notice and was applied at renewal.

04 · Same price, different economics

More usage on cheaper models. Less on expensive ones.

Metering by cost is not hostile in itself. AI models have real, variable costs, and Cursor said “the hardest requests cost an order of magnitude more than simple ones”.12 The effect depends on which model a customer uses.

Read with these caveats

  • Median, not every user. The figures describe a median request. Long agent tasks can cost far more.
  • Different units. An old request was one prompt of any size; the new figure is a median-cost request.
  • Sonnet 4 was new. It had launched about three weeks earlier and was billed at a temporary discount of half a request.2021
  • No slow fallback for Sonnet 4. It was already outside the unlimited slow pool.18
  • Bonus usage not counted. Cursor’s docs later promised “additional bonus usage” but never quantified it.17

The new system could provide more usage on cheaper models and less on expensive ones.CHI reading of Cursor’s published figures

Cursor staff said the same in August 2025: the change “does mean that heavy users of the most expensive models, like Claude 4 Sonnet, may get less than 500 requests each month”, while users “can also now get more than 500 requests in a month from cheaper models”.22

~$44Allowance used by 500 median Sonnet 4 requests
$20Included in Pro

Illustration only — our calculation from Cursor’s published figures. Cursor said $20 covered about 225 median Sonnet 4 requests, so 500 such requests would consume about $44 of allowance at API prices ($20 ÷ 225 × 500 = $44.44). This is not a bill any customer was shown and not Cursor’s own cost. Cheaper models could exceed the old 500-request count; median usage is not every user’s usage; and bonus usage is not included in the arithmetic.

How old requests were counted. Under Cursor’s documentation, an ordinary Claude Sonnet request counted as one request and thinking mode as two.181921 Cursor’s 4 July 2025 post described the old plan “with Sonnet models costing two requests”, which does not match its own docs for ordinary use.12

05 · Disclosure

The number disappeared before the subscription did.

Cursor still sells the plan, but no longer publishes the exact quantity inside it. A prospective customer can see the price and a multiple. Not the amount.

  1. To Jun 2025 500 requests A fixed count

    “500 requests per month”, on Cursor’s pricing page.4

  2. From Jul 2025 At least $20 of model inference Measurable, in dollars

    “at least $20 of model inference at API prices per month”.10

  3. Sep 2025 – 24 Aug 2026 $20 · $70 · $400 Pro · Pro+ · Ultra

    Cursor’s docs listed the third-party model usage included in each individual plan.1723

  4. From 29 Aug 2026 Included · 3x · 20x · 5x No base amount stated

    The docs table reads “Included”; plans are described as multiples of an unstated base.2426

PlanCurrent wordingExact quantity published?
Pro · $20Current wording“Extended limits on Agent”; docs: “Included”2526Exact quantity published?No
Pro+ · $60Current wording“3x Pro limits on Agent”Exact quantity published?No, a multiple only
Ultra · $200Current wording“20x Pro limits on Agent”Exact quantity published?No, a multiple only
Teams Standard · $40Current wording“the standard Teams usage allowance”30Exact quantity published?No
Teams Premium · $120Current wording“5x the usage of a Standard seat”2930Exact quantity published?No, a multiple only

“We no longer publish specific dollar amounts for included usage, since it can vary by plan and change over time.”Cursor staff, Cursor forum, 25 September 202627

The dollar figures were in Cursor’s documentation in an archived copy of 24 August 2026 and absent from one of 29 August 2026; no announcement of the change was found.2324 Subscribers see the share they have used as a percentage.27 This shows a withdrawal of disclosure, not a proven reduction: no evidence establishes that any allowance shrank.

06 · The legacy plan

Legacy stayed alive, but stopped moving forward.

Some subscribers kept the old 500-request plan. It still works for the models it covered. New frontier models increasingly sit outside it, behind Max Mode, which is billed at “the model’s API rate plus 20%”.36

17 March 2026 · Cursor staff
“This change does not apply to individual plans” … “Are individual plans affected? No, with the exception of GPT 5.4, which has been Max Mode only for all users since launch. The March 16th change does not impact individual plans.”
Forum post announcing Max Mode for frontier models on legacy Team and Enterprise plans31
7 July 2026 · Cursor staff
“Since the release of GPT-5.4 in March, new frontier models have required Max Mode for users on legacy request-based pricing. A bug meant this wasn’t always being enforced. That’s now fixed”
Forum reply in a thread opened by individual legacy subscribers32
The two statements are in tension, not a flat contradiction.

Cursor staff explained that the March post covered moving existing models to Max Mode for teams, while a separate going-forward rule, stated in posts on new-model launches from March, applied to new frontier models on individual plans.3337 Staff said the rule had been enforced for most legacy accounts since March and that “A subset of accounts had been running on the older behavior longer than intended”.33

What legacy users kept

Request-based pricing “remains available for legacy Pro plans”, including models such as Opus 4.6, GPT-5.3 and Composer 2.5.34

Cursor did not promise that future frontier models would stay inside the 500-request allowance. Staff now say: “As a general expectation, you can assume new models will require Max mode.”34

We found no staff statement that affected individual subscribers were notified before the July enforcement.

What moves an account off legacy

Cursor staff named three events, and said the move is irreversible:35

  1. Upgrading to Pro+ or Ultra
  2. Changing the billing cycle, monthly to annual or back, through the new checkout
  3. Cancelling and subscribing again

“The opt-out option that used to let you stay on legacy isn’t available anymore”, staff added, and “I can’t promise legacy pricing will stay forever.”35

07 · Billing control

Metered does not mean uncontrollable.

This is important counterevidence. An individual subscriber cannot be charged beyond the subscription unless they switch paid usage on. Individual and team plans work differently and are shown separately.

Individual plans

  • Overage is opt-in“Individual plans: On-demand usage must be explicitly enabled in your settings.”38
  • Spend limits existCursor’s help pages describe monthly spending limits for on-demand usage.39
  • Requests stop when the allowance runs outWith on-demand off, Cursor’s help page says it will “stop requests once your included usage runs out.”38
  • Brief overshoot is creditedUsage that briefly exceeds a limit is credited “as a temporary spend-limit credit”, though “Raising a limit mid-cycle can make previously credited overage billable.”39

Teams plans

  • Overage is on by default“On-demand usage is enabled by default for the Teams plan.”30
  • Team-wide limits are available“Teams can configure monthly team-wide spending limits.”30
  • Per-member limits are Enterprise-only“Per-member spend limits are available on Enterprise plans.”30
  • A token rate on third-party modelsTeams pay a Cursor Token Rate of $0.25 per million tokens on requests to third-party models.30

Visibility. An in-editor bar shows the percentage of included usage used. In July 2026 a staff member wrote that Cursor does not “currently have mid-cycle warnings like “you’ve used X%” for Pro Individual”.40 The behaviour described here is Cursor’s stated policy; it was not tested in the product.

08 · The counterweight

What stayed favourable.

This side of the ledger is why the verdict is PARTIALLY SUPPORTS, not SUPPORTS or STRONGLY SUPPORTS.

Headline price

Pro $20, Teams $40

Pro was $20 a month on every archived Cursor pricing page we opened from May 2023, and Business/Teams $40 a seat from September 2023. Both are unchanged today.1326

Cheaper models

More than 500 requests’ worth

By Cursor’s median examples, $20 covered about 550 Gemini or 650 GPT-4.1 requests, more than the old count.12

Overage

Opt-in for individuals

Paid usage beyond the plan must be switched on, which sharply limits surprise billing.38

Remedy

Apology and refunds

Eighteen days after the June 2025 change Cursor apologised and offered refunds for unexpected usage.12

Grandfathering

Auto change applied at renewal

Annual subscribers kept unlimited Auto until their annual renewal.16

New functionality

Added, not taken away

Max Mode and background agents were new options, not entitlements previously included in Pro.441 The product became substantially more capable over the period.

Cursor’s stated reason for the 2025 change: “New models can spend more tokens per request on longer-horizon tasks. Though most users’ costs have stayed fairly constant, the hardest requests cost an order of magnitude more than simple ones.”12

09 · Pattern classification

One pattern in specific cases. One partial. One narrow.

CHI’s institutional signal taxonomy, applied without forcing a fit and without inferring intent.

SP001

Unmonetized Access / Entitlement Monetization

Supported in specific instances

Unlimited slow requests ended for users moved to the new plan; unlimited Auto was later metered; new frontier models on legacy plans moved outside the 500-request allowance into Max Mode.

Caveats: slow access carried a throttling warning; Sonnet 4 was already outside the slow pool; future models were never promised to the legacy plan; the replacement plan included at least its price in model usage.

SP002

Existing-Customer Yield Optimization

Partially supported

Users of expensive models could receive less included usage at the same price; heavy users face more variable cost; the risk of an expensive request moved toward the customer.

Counterweight: cheaper-model users could receive more; individual overage is opt-in; the headline price stayed flat; refunds were offered; no evidence shows typical users pay more.

SP003

Legacy Product Deprioritization / Portfolio Concentration

Supported · narrow

The opt-back was removed; new frontier models require Max Mode on the legacy plan; three account actions move users off it irreversibly; its future is not guaranteed.

But: legacy users kept the models it already covered; the number of legacy users is unknown; Cursor never promised indefinite support for new models.

SP004

Regulated Pass-through / Customer Recovery

Not applicable

No regulatory or externally mandated cost was passed to customers. The Max Mode surcharge and team token rate are commercial charges.

10 · What changed

Worse, better, and unknown.

What got worse

  • A fixed request entitlement became variable, model-cost consumption.
  • The unlimited slow fallback disappeared from the new plan.
  • Unlimited Auto was later metered.
  • Included-dollar figures were removed from current documentation.
  • Legacy plans stopped carrying forward to new frontier models.

What got better or stayed favourable

  • Pro stayed $20.
  • Teams stayed $40.
  • Cheaper-model users could receive more usage.
  • Individual overage remains opt-in.
  • Spend limits exist.
  • Cursor offered refunds after the June 2025 confusion.
  • Annual Auto terms were kept until renewal.
  • Functionality expanded substantially.

What remains unknown

  • The current included amount for Pro, Pro+ and Ultra.
  • The size of the “bonus usage” Cursor described.
  • Whether every annual subscriber got identical grandfathering in practice.
  • How many users were materially worse off.
  • How many users received refunds.

Final CHI assessment · Partially supports

The subscription became more conditional, more variable and less transparent.

Cursor did not raise the $20 headline price. Instead, it changed the unit underneath it — from a fixed request allowance with an unlimited fallback, to model-cost metering, then to allowances whose exact size is no longer published.

That does not mean every customer lost value. Cursor’s own figures show cheaper models could provide more usage, individual overage remained opt-in, and the company added substantial new capability.

The evidence therefore partially supports customer hostility: the strongest case is not that Cursor became more expensive for everyone, but that the subscription became more conditional, more variable, and less transparent over time. No numerical CHI, CVI or CFS score has been assigned.

Pro price raisedNo
What $20 includesBecame variable
Included amount publishedNo
Cheaper-model usageMore
Individual overageOpt-in
CHI conclusionPARTIALLY SUPPORTS

The price didn’t change.
How much it promised did.

Sources

Evidence behind this page.

Every statement on this page comes from the Cursor CHI research dossier and its publication verification (both 11 October 2026), and traces to a source below. Quotation marks are used only for wording checked against the original or an archived copy of it. Staff means a post by an account flagged as Cursor staff on Cursor’s forum; archived means a Wayback Machine copy of a Cursor page; calculated means arithmetic from Cursor’s published figures.

  1. Cursor pricing page, archived 13 May 2023. Internet Archive
  2. Cursor pricing page, archived 13 Jul 2023. Internet Archive
  3. Cursor pricing page, archived 4 Sep 2023 (first capture showing Business at $40). Internet Archive
  4. Cursor pricing page, archived 11 Jun 2025. Internet Archive
  5. Cursor pricing page, archived 16 Jun 2025, 17:32 UTC (old plan). Internet Archive
  6. Cursor pricing page, archived 17 Jun 2025, 07:42 UTC (new plan). Internet Archive
  7. Cursor blog, “Introducing the Ultra Plan”, dated 16 Jun 2025, archived 17 Jun 2025. Internet Archive
  8. Cursor docs, “Rate limits”, archived 18 Jun 2025. Internet Archive
  9. Cursor forum, pricing megathread, staff reply (post 64). forum.cursor.com Staff
  10. Cursor blog, same post after the edit Cursor dates to 30 Jun 2025, archived 3 Jul 2025. Internet Archive
  11. Cursor pricing page, archived 2 Jul 2025. Internet Archive
  12. Cursor blog, “Clarifying Our Pricing”, 4 Jul 2025, archived 5 Jul 2025. Internet Archive
  13. Cursor forum, staff reply on the opt-out, 10 Jul 2025. forum.cursor.com Staff
  14. Cursor forum, “Why the opt-out button is no longer available”, staff reply 14 Jul 2025. forum.cursor.com Staff
  15. Cursor forum, “Auto mode now charged?”, staff reply 22 Jul 2025. forum.cursor.com Staff
  16. Cursor blog, “Updates to Teams and Auto”, 12 Aug 2025, archived 14 Aug 2025. Internet Archive
  17. Cursor docs, “Pricing”, archived 5 Sep 2025. Internet Archive
  18. Cursor docs, “Models & Pricing”, archived 21 May 2025 (request costs by model). Internet Archive
  19. Cursor changelog 0.47, March 2025 (thinking mode at two requests). cursor.com
  20. Cursor forum, “Claude 4 Sonnet & Opus now in Cursor”, staff post 22 May 2025. forum.cursor.com Staff
  21. Cursor forum, “Claude 4 – End of Discounted Rate”, staff post 19 Jun 2025. forum.cursor.com Staff
  22. Cursor forum, “Cursor is expensive”, staff reply 6 Aug 2025. forum.cursor.com Staff
  23. Cursor docs, “Models & Pricing”, archived 24 Aug 2026 (dollar figures present). Internet Archive
  24. Cursor docs, “Models & Pricing”, archived 29 Aug 2026 (dollar figures absent). Internet Archive
  25. Cursor docs, “Models & Pricing” (read 11 Oct 2026). cursor.com
  26. Cursor pricing page (read 11 Oct 2026). cursor.com
  27. Cursor forum, usage-meter thread, staff reply 25 Sep 2026. forum.cursor.com Staff
  28. Cursor blog, 11 Feb 2026 (still reads “at least $20”). cursor.com
  29. Cursor blog, Teams pricing, 1 Jun 2026. cursor.com
  30. Cursor docs, “Team pricing” (read 11 Oct 2026). cursor.com
  31. Cursor forum, “Frontier Models Moving to Max Mode for Legacy Team/Enterprise Plans”, staff post 17 Mar 2026 (archived 17 Mar 2026). forum.cursor.com Staff
  32. Cursor forum, “Legacy individual plan: Max Mode required”, staff reply 7 Jul 2026 (post 7). forum.cursor.com Staff
  33. Same thread, staff reply 7 Jul 2026 (post 205). forum.cursor.com Staff
  34. Same thread, staff reply 10 Jul 2026 (post 597). forum.cursor.com Staff
  35. Cursor forum, “What happens to legacy request-based plans after renewal”, staff reply 5 Aug 2026. forum.cursor.com Staff
  36. Cursor help, “Max Mode on legacy plans” (read 11 Oct 2026). cursor.com
  37. Cursor forum, GPT-5.4 launch thread, staff post 5 Mar 2026. forum.cursor.com Staff
  38. Cursor help, “Overages” (read 11 Oct 2026). cursor.com
  39. Cursor help, “Spend limits” (read 11 Oct 2026). cursor.com
  40. Cursor forum, usage-warnings request, staff reply 21 Jul 2026. forum.cursor.com Staff
  41. Cursor changelog 0.50, 15 May 2025 (Max mode token pricing; Background Agent preview). cursor.com

Limits carried onto this page. Billing behaviour is described from Cursor’s published policy and staff statements; the product itself was not tested. Whether subscribers were emailed before the June 2025 change could not be verified. How many customers had unexpected charges, were refunded, or remain on legacy plans is not public. The current size of any plan’s included usage is not published, and the “bonus usage” Cursor described was never quantified. Cursor’s later wording says Auto now bills at the list price of the model it routes to; the date that change took effect has not been established from a Cursor announcement, so no date is given here. User reconstructions of current allowances, press reports about Cursor’s finances and its ownership are not used.

Status. Research frozen 11 October 2026 and publication-verified the same day. No numerical CHI, CVI or CFS score has been assigned; scoring is reserved for a later calibration stage.

Reading this beside the rest of the index

AI subscriptions, and what the price buys.

Lovable shows a stable price while the unit of sale was redefined. Cursor shows a stable price while the included amount became variable and then unpublished. Read the methodology for why no score is assigned yet.

Same sector · AIRead Lovable → Company comparison · AI subscriptionsRead OpenAI vs Anthropic → How CHI scores, and why no score is assigned hereRead the methodology → Every company in the sectorBrowse AI →