The X Subscription Bundles Cursor With Grok. Third-Party Models Now Bill Separately.

The X Subscription replaces the old SuperGrok, Cursor and X Premium plans with five tiers: Free, Plus at $8, Premium at $30, Super at $100 and Ultra at $200. Paid tiers share one usage pool across Cursor and the Grok products. Multipliers are 4x, 15x and 50x against Plus.

Here’s what the coverage buried. Cursor’s own pricing page says third-party model usage “is not included in the monthly prices shown and must be purchased separately.” Claude and GPT stay available at API list price, billed on top. The old Pro plan at $20, which included $20 of third-party usage, isn’t listed anywhere on that page anymore.

So a developer who paid $20 and spent $20 on Claude is looking at $30 plus that usage on top. Cursor says existing plans continue for now, with migration “with advance notice” and no timeline given.

Best for anyone happy running Grok models all day. Not ideal for anyone whose daily driver is Claude or GPT inside Cursor.

Somebody is about to open their Cursor bill and find a number they didn’t agree to.

Not this month. Probably not next month either. The announcement went out as a bundle with a lower entry price. $8, where three separate subscriptions used to cost $90 together. Every writeup led with that. The rest of it sits on Cursor’s own pricing page in plain language. It changes what the tool costs for anyone who uses it the way most people do.

We read the pricing page rather than the press coverage. Here’s the arithmetic.

What It Actually Is

Tier names, prices and multipliers came off cursor.com/pricing on October 13, 2026. Two outlets reported the same five numbers. The before-and-after figures for the old Cursor plans come from The New Stack.

One thing we’re not going to assert: who exactly is selling this. The announcement came from an account called SpaceX Super Intelligence. Outlets covering it describe the seller variously as X, as xAI and as SpaceX. The pricing page is Cursor’s. We’ll stick to what each page says.

The Old Pro Plan Is Gone From The Page

Open cursor.com/pricing today. The ladder reads Free, Plus at $8, Premium at $30, Super at $100, Ultra at $200, then Teams and Enterprise. There’s no Pro tier at $20. There’s no Pro+ at $60 either.

That matters because of what those plans used to include. Pro at $20 came with $20 of third-party model usage at API prices. Pro+ at $60 came with $70 of it. The old Ultra at $200 came with $400 of it. You paid a flat monthly number and a chunk of your Claude or GPT usage was already covered inside it.

The new ladder works differently and Cursor says so directly.

One Sentence Does All The Work

Four lines, quoted exactly from Cursor’s pricing page:

“Third Party model usage in Cursor is not included in the monthly prices shown and must be purchased separately.”

“Premium and above include generous usage of Grok models but do not include an allowance for third-party models.”

“Third-party models remain available in Cursor at API list price.”

And on the Premium, Super and Ultra cards: “3rd party models are available in Cursor, billed separately by usage.”

Read those four together and the shape is clear. The subscription buys you Grok usage. Everything else is a pass-through at list price, from the first request, at every tier including the $200 one. The page calls the mechanism “top-offs” rather than credits, which is worth knowing if you go looking for the setting.

Plus at $8 doesn’t include third-party models at all. Those start at Premium. So $30 is the floor for using Claude inside Cursor. At that floor the usage still bills extra.

What A Cursor Pro User Actually Pays Now

Do this one on paper. The headline and the bill point in opposite directions.

The $20 User Who Spent Their Allowance

Say you were on Pro at $20. You used the full $20 of included third-party usage every month, which is normal for anyone running Claude on a real codebase. Your total was $20.

To keep the same models you move to Premium at $30. The usage is no longer included, so the same $20 of Claude sits on top. You’re at $50 for the month doing exactly what you did before. The New Stack ran the same arithmetic and landed on the same figure.

That’s a 150% increase on an announcement that read as a price cut.

The Pro+ And Old Ultra Users

Pro+ at $60 included $70 of third-party usage. If you were using most of that, call it $130 of value for $60. Moving to Super at $100 gets you 15x the Plus Grok allowance and none of that coverage. The same workload costs $100 plus whatever the models bill.

Old Ultra at $200 included $400 of third-party usage, which was the most generous ratio on the ladder. The new Ultra is the same $200 with a 50x Grok multiplier and third-party billed separately. If your $400 was going to Claude, that’s the biggest swing of anyone.

The User Who Was Already On Grok

If Grok was already your model, almost none of this is bad news. You get a bigger usage pool shared across Cursor and the Grok products. The entry point drops to $8. The vendor says each tier gives more usage per dollar than the plan it replaces. Teams get a better deal too, $30 per seat against $40 before.

It’s worth being clear about how real that is, because the Grok case gets lost whenever anyone writes about the Claude case. A solo developer who was paying $20 for Cursor and happily using Grok inside it now pays $8 for the same thing with a larger pool behind it. Somebody who was carrying SuperGrok alongside Cursor alongside X Premium was paying roughly $90 across three bills and now pays $30 for one. That isn’t a repackaging trick. That’s a genuine consolidation and it’s the group the announcement was written for.

The trouble is that most Cursor users sit outside that group. Claude and GPT have been the default choice inside Cursor for most of the tool’s life, which is exactly why the included third-party allowance existed in the first place. Removing it lands hardest on the people who were using the product as designed.

So the honest version isn’t “prices went up.” The pricing moved to favour one model family. Your bill follows whichever one you actually use.

Cloud Agents Moved Up A Tier

Two features changed tiers rather than price. That sort of change never makes a headline.

Cloud agents and Origin used to sit inside the $20 Pro plan. They’re now Premium and above, so $30. Anyone who was on Pro specifically for cloud agents pays $10 more a month to keep them, before any model usage.

The Premium card lists automations alongside cloud agents. Super adds priority model access at peak times. Ultra adds earliest access to new products and models. None of those were separately priced before, so there’s nothing to compare them against. Treat them as new tier bait rather than as something taken away.

What Happens To The Plan You’re On

Nothing, for now and the words matter.

A Cursor support engineer said existing plans continue “for now” at the same price and the same features. Asked what happens after that, the answer: users who don’t migrate “will, sometime in the future, be migrated with advance notice.” Another response said there’s no timeline for moving the remaining customers.

The vendor side says it shorter: “Nothing changes for existing SuperGrok, Cursor, or X Premium subscribers.” Switching early gets you a prorated refund.

So you’re not being moved today. You are being told, in writing, that you will be moved eventually. The notice comes at an unspecified point. The plan you’re on is no longer sold. That’s a grandfathered tier and grandfathered tiers have a shelf life.

We’ve watched this sequence run at more than one vendor already. A plan stops being sold. Existing customers are told nothing changes. Some months later a migration notice arrives with a date on it and the old terms don’t survive the move. Nobody involved is being dishonest at any step. The language is accurate every time it’s used, which is why reading it as a promise rather than as a status report is the mistake worth avoiding. If you’re on old Pro and Claude is your model, that clock started October 12.

Our Qwen piece covered this pattern at another vendor. What’s generous while a product wins users becomes a business decision once it has them. Nothing here is sneaky. It’s a different deal than the one people signed up for, announced in language that emphasised the cheaper part.

What The Vendor Claims And What We Can Check

Three claims can’t be checked from a pricing page. We’re flagging them rather than repeating them as fact.

“Each comparable new tier provides more usage-per-dollar than the plan it replaces.” True in Grok tokens. False in dollars for anyone whose usage was third-party. Both can hold at once, which is why the sentence works.

Every paid plan gives “at least twice the coding usage” with their own models compared to the matching Cursor plan. Nobody outside can measure that. The old plans metered third-party usage in dollars. The new ones meter Grok usage in multiples of a Plus allowance whose size isn’t published anywhere we could find. There’s no shared unit to compare.

Premium covers what SuperGrok, Cursor Pro and X Premium+ cost separately, which one outlet put at $90 a month combined. We didn’t verify those three historical prices ourselves, so that one carries their byline and not ours.

Why The Coverage Led With $8

Nearly every writeup opened with the same framing. Three subscriptions used to cost around $90 together. The entry tier is now $8. That’s a real number off a real page, so it isn’t wrong.

The reason it dominated is simpler than bias. Announcement posts are written to be quotable and pricing pages are written to be accurate. The announcement carried the tier list. The exclusion sentence lives on cursor.com/pricing, under a heading about third-party models, below the comparison grid. A writer working from the post would never see it.

So the pattern to take from this isn’t about one vendor. It’s that the number in the headline and the number on your card come from two different documents. One is marketing copy and the other is a terms page. When they disagree, the terms page is the one that bills you.

We’d rather flag that than imply anybody hid something. The sentence is published in plain English on a public page. It just wasn’t the sentence anyone chose to lead with.

The Workaround People Will Reach For

One tool at the top of our GitHub scan this week exists for exactly this situation. Either good timing or a sign of where this is heading.

magpie took 1,104 stars yesterday alone and it’s 11,354 total at 19 days old, MIT licensed. What it does: points any agent at any provider from the menu bar. Claude Code on Kimi, Codex on DeepSeek, Gemini CLI on GLM. One local gateway serves them all. When a quota runs out it moves to the next account without the agent seeing an error.

That’s the structural answer to model usage leaving a subscription. If the tool you pay for stops covering the model you want, you route the model yourself and pay the provider directly. Whether that’s cheaper depends on your volume. We haven’t measured it, so this isn’t a recommendation yet. It’s on the list to install and test properly.

Note that magpie carries sponsor logos in its README and has a hosted site. Check whether a paid tier exists before calling it free. We haven’t, which is why this isn’t a review.

Our Make vs Zapier vs n8n comparison does the cost-per-run arithmetic on workflow tools. Same exercise, different layer.

What A Two Person Team Pays

Team seats went from $40 to $30, which reads like a straight win and mostly is. Two seats used to run $80 a month. The same two seats now run $60. That’s $240 a year back for a small shop, before anything else changes.

Then the same question arrives. What models does the team actually use? If both developers run Grok, the $60 is the whole bill and the shared pool is bigger than what $80 bought. If both run Claude, the $60 is the subscription and the model usage sits on top at list price, metered per request rather than capped by an allowance.

That second case is worth modelling before anyone migrates. A team on two old Pro seats had $40 of third-party usage included between them. On the new Teams plan that $40 becomes a variable line item on the card. In a quiet month it costs less. In a month where somebody points an agent at a large repo and lets it run, it doesn’t.

The practical move for a team is to pull last month’s actual third-party spend before deciding anything. Cursor meters it, so the number exists. If it came in under $20 a seat, the new Teams price is cheaper in total. If it ran over, the seat discount gets eaten and then some. That’s a five minute check that settles the whole question for your specific team rather than for a hypothetical one.

Because the usage is pooled across seats on paid plans, one heavy user also drains the allowance the other one was counting on. That’s not new behaviour for shared pools, but it’s new to anyone coming off individual plans where the allowance was per person.

Verified vs Asserted

What To Watch Next

Three things will decide whether this ages into a good deal or a bad one.

First, the size of the Plus allowance. Every multiplier on the ladder is expressed against it, so a 50x multiplier means nothing in isolation. Until that base number is published, the usage comparisons between old plans and new ones can’t be checked by anyone outside the company.

Second, the migration notice. Existing plans continue “for now” with no date attached. When that notice does land, the question to ask is whether it carries the old included usage forward for a grace period or whether it moves everyone onto the pass-through model at once.

Third, whether third-party models stay available at all. Right now the page says they remain in Cursor at API list price, which is a commitment to access rather than to cost. A vendor that wants its own models used has an obvious next lever. Removing a model is a much louder change than repricing one, so we’d expect it to come with warning. Still, the lever exists.

None of those three are predictions. They’re the specific things we’ll check when we revisit this. They’re also the things worth checking yourself before committing a team to a tier.

What You Should Actually Do

If You’re On Old Cursor Pro

Stay put and start watching. Your plan still works at the old price with the old included usage. Migration needs advance notice that hasn’t arrived. Meanwhile, open your usage page and write down what you actually spent on third-party models last month. That number plus $30 is your real cost after migration and it’s the figure to compare any alternative against.

If You’re Signing Up Today

Premium at $30 is the floor if Claude or GPT is your model. Budget the usage separately on top. Plus at $8 only makes sense if you’re happy on Grok, since third-party models aren’t available there at all.

If Your Model Is Claude Specifically

Price the direct route. Claude through its own subscription plus a router is a real comparison now rather than a weird one. Our Claude Pro review covers what that tier includes. The answer depends on your volume, so do it with your own numbers rather than ours.

The Habit Worth Building

Screenshot your pricing page. Every tool in this stack has changed its terms this year. The people who notice early are the ones with a record of what the deal was. A screenshot with a date on it has settled more billing arguments than any amount of indignation.

The Part Worth Keeping

The announcement wasn’t dishonest. Five tiers, a lower entry price, cheaper team seats and one shared usage pool are all real. For a developer who lives on Grok models this beats what came before.

What changed underneath is who pays for the models. Usage that used to arrive inside a flat monthly price is a pass-through at list price now. The plan that bundled it isn’t sold anymore. If your model is Claude, your bill goes up while the sticker price goes down. Both sentences are true at once. That’s the whole story and it took reading the pricing page instead of the announcement to find it. /separator

Charts and Blocks

The Ladder As The Page Lists It

What The Same Month Costs Before And After