Cursor's $20 plan is gone — and the models most Cursor users run left the usage pool

On 12 October 2026 SpaceXAI collapsed three products into one price list. Grok, Cursor and X Premium are now sold as The X Subscription, a five-rung ladder: Free, Plus at $8, Premium at $30, Super at $100 and Ultra at $200 a month, with Teams at $30+ per user. The company’s framing was simple enough — “We’ve unified plans across Grok, Cursor, and X Premium.”

Read as a price list, this is one of the largest cuts of the year. $8 a month now covers the Cursor editor, Grok Chat, Voice, Imagine, Build, the Grok API and X Premium’s checkmark and reduced ads. Before this, X Premium+ alone was $40, SuperGrok alone was $30, and Cursor Pro alone was $20.

Read as a billing model, something different happened, and it is the part worth your attention: the shared usage pool is Grok-only.

What actually moved

Cursor’s pricing page is explicit on both halves. On Plus: “Only Grok models are available on this plan.” On Premium and above: “3rd party models are available in Cursor, billed separately by usage.”

That second sentence is the whole story. Claude and GPT have not been removed from Cursor. They have been moved out of the subscription and onto a meter, billed at API list prices through top-ups. The usage pool you pay $30, $100 or $200 a month for covers Grok models and nothing else.

Here is the old ladder against the new one, as both stood on 13 October 2026:

There is no $20 rung any more. A developer on Cursor Pro who uses Claude in the editor has two moves: down to Plus at $8 and give up every non-Grok model, or up to Premium at $30 and start paying per token for the models they were already using.

The arithmetic the pricing page does not do

Take a developer whose daily driver in Cursor is Claude Sonnet 5.5, at its published list rate of $2 input / $0.10 cached input / $10 output per million tokens. A heavy agentic day — call it 3M tokens of cached repository prefix, 200K of fresh input and 150K of output — costs about $2.20. Across 21 working days, roughly $46 a month.

So that developer’s bill goes from a flat $20 to $30 plus about $46, and the second number is an estimate rather than a cap. Run the same day on Claude Fable 5.1 at $10/$50 and the metered half multiplies. Run it on Grok 4.7 — $2/$6 per million, and inside the pool — and the marginal cost is zero.

That gap is not an accident of rate cards. It is the entire design. The new ladder prices Grok usage at a flat subscription and everything else at list, which makes the house model the only one whose cost you can predict.

The number nobody can check

The old Cursor ladder never published a credit allowance, which this publication has flagged before. The new one does not fix that; it formalises it.

SpaceXAI publishes relative values only: Premium is “4x more usage compared to Plus,” Super is 15x, Ultra is 50x. The absolute size of a Plus allowance appears nowhere. So a buyer cannot answer the one question that decides the purchase — will Premium’s pool cover my month? — before paying for a month and finding out.

Combine the two halves and the knowable portion of a Cursor bill is now the subscription fee alone. The pool has no published size, and the models outside it have no published ceiling.

The team tier is where this gets expensive

The individual tiers are the easy part of the story. The team numbers are where a finance department should look twice.

Cursor’s old team ladder was Teams Standard at $40/user and Teams Premium at $120/user, both flat, both including third-party model usage. The new one is Teams from $30/user, sized at 4–50x the Plus pool. On the headline that is a 25% cut on the entry seat.

It is not, for most teams. A twenty-developer team standardised on Claude inside Cursor previously paid $800 a month, fully predictable. On the new ladder it pays $600 a month plus every Claude token at list — and using the same per-developer estimate above, roughly $46 a seat, that is about $920 a month of metered spend on top, for a total near $1,520. The subscription got cheaper and the bill nearly doubled.

SpaceXAI does ship the control that matters here: Teams and Enterprise carry spend controls, and Enterprise adds granular ones. But a spend control is a cap on damage, not a forecast. The planning problem is that a team cannot build next quarter’s budget from the pricing page, because the one quantity that drives the bill — how many third-party tokens twenty developers consume in a month of agentic work — is the quantity nobody has measured, precisely because it was previously free.

The practical move for any team on a legacy Cursor plan is to turn on whatever per-model usage reporting exists and collect three months of data before the migration arrives. That number is currently costing you nothing to produce and will be the most valuable input you have when it stops being free.

This is the second half of a move that started in August

In August, OpenAI set 12 November 2026 as the date it withdraws model access from Cursor — the event that established multi-model access as a contract rather than a feature. That removed one vendor from the menu by supplier decision.

This removes the rest by pricing. SpaceX acquired Anysphere in June 2026 and has shipped steadily toward integration since: Grok 4.6 as the Cursor default in August, Grok 4.7 in September, and now a single subscription where Cursor is one surface among Chat, Voice, Imagine and Build. An editor that was sold on model neutrality is now the coding surface of a model vendor, and its pricing says so plainly.

Whether that is bad depends entirely on which model you want. SpaceXAI has made Grok-based development remarkably cheap — $8 for an editor, a frontier-ish model, an API and a social subscription is a real number, and for a solo developer committed to Grok it is the best value on this page. It has simultaneously made Cursor the most expensive way to run Claude or GPT, because every competitor bundles its own model for a flat fee.

What this does to the buying decision

The comparison that mattered a week ago was Cursor Pro at $20 against Claude Pro at $20 and ChatGPT Plus at $20 — three flat fees, three different philosophies, model cost included in all three.

That symmetry is gone. ChatGPT Plus still bundles Codex with GPT-6.1 Sol at no marginal cost, and Claude Pro still bundles Claude Code. Cursor now bundles Grok, and charges list for the rest. For a buyer whose preference is Claude or GPT, the flat-fee options are the two vendor-owned agents, and the IDE is the metered one. Our Cursor vs Claude Code comparison carries the longer version of that decision.

For existing subscribers there is no urgency and no deadline: legacy plans continue, only closed to new signups, with the company saying migration will happen “eventually” and that there is no timeline. The correct reading of that is a grandfathered rate with an unannounced expiry. Use the time to measure what your third-party token spend actually is, because on the day the migration lands that number stops being free.

What changed in our trackers

pricing.json now carries the new Cursor ladder with the legacy plans retained and marked closed to new signups, plus a dated launch event for 12 October 2026. The pricing tracker shows the before-and-after.

Questions or corrections? Email Pick Right. Want the full list? See all news.