The X Subscription Bundles Grok and Cursor: The Hidden Cost

On October 12, 2026, SpaceXAI posted a single line on X that reorganized how millions of people pay for AI: “We’ve unified plans across Grok, Cursor, and X Premium. Introducing The X Subscription.” Five tiers, from free to $200 a month. On paper, the price of the most popular combination fell from roughly $90 to $30.

That headline is real. It is also incomplete. Buried in the same announcement, and spelled out in fine print on Cursor’s own pricing page, is the clause that decides whether the new bundle saves you money or costs you more: third-party models in Cursor move to API list price, billed separately by usage, outside the monthly fee.

For a light user, the bundle is a genuine discount. For a heavy user, the same change quietly removes the flat-rate allowance that made the old subscription predictable. This is the pattern tokenkarma tracks across every provider, and this week it arrived in one of the largest AI subscriptions on the market.

What The X Subscription actually includes

The bundle collapses three previously separate products into one bill:

• Free ($0): daily Grok access, a basic allowance for Cursor coding agents, standard X.
• Plus ($8): more AI usage, a checkmark, creator tools, fewer ads. Only Grok models are available on this plan.
• Premium ($30): the most popular tier, with 4x the Plus usage, cloud agents and automations, code hosting, AI code reviews, and Grok Bot.
• Super ($100): 15x the Plus usage plus priority model access at peak times.
• Ultra ($200): 50x the Plus usage and earliest access to new products and models.

SpaceXAI states that SuperGrok, Cursor Pro and X Premium+ previously cost about $90 a month combined, and that Premium now covers all three. It also claims each comparable tier delivers more usage-per-dollar than the plan it replaces. That math only holds if the cost of serving each request falls faster than the price, which is why the third-party clause exists at all.

The clause that changes your bill

Here is the part heavy users need to read twice. On every paid tier from Premium upward, Cursor now lists “Third-party models at API list price.” The footnote is explicit: third-party model usage in Cursor is not included in the monthly price and must be purchased separately. Only Grok models are covered by the flat allowance.

In practice, this means the flat subscription now pays for Grok and for X, while every call you make to a competing frontier model through Cursor is metered per request. If your workflow is mostly Grok, nothing changes and you pay less. If your workflow depends on running Claude, GPT, or Gemini inside Cursor to compare outputs or to route sensitive work, your effective monthly cost is now the subscription plus an open-ended API tab.

SpaceXAI has not published the per-request rates, nor has it said how many requests or tokens the Plus tier allows. Days after launch, the entry paid tier still has no published usage ceiling. That opacity is itself a budgeting risk: you cannot forecast a bill on a number nobody has stated.

Why the usage pool matters more than the price

All paid tiers “share one usage pool across Cursor, Grok Build, and other Grok products.” One pool sounds convenient. It also means every experiment, every long agent run, and every casual generation drains the same bucket that funds your actual coding work. The old world had separate allowances that failed independently. The new world has a single fuse.

Two consequences follow for anyone spending real money:

1. Peak-time priority is now a paid feature. Super and Ultra get “priority model access at peak times.” On a shared pool, priority is how the provider allocates scarce capacity when everyone is working. If you are on Premium and the pool is saturated, your agent slows down exactly when you need it most.
2. Auto-routing blurs the line between cheap and expensive. When a plan advertises one pool and many products, you lose the visible signal of which action consumed how much. This is the same blurring that Anthropic introduced when it merged chat and Cowork into one interface, and that Google introduced when it routed free Gemini users to Flash-Lite. The industry is converging on a single answer: make the meter invisible, then sell more of it.

How it compares to the rest of the market

The X Subscription is not happening in a vacuum. Three parallel moves landed within days, and together they redraw the map for heavy users:

• Google made Gemini 4 Argon broadly available on October 13 with limited-time introductory API pricing that doubles on December 1, 2026. The cheap number is a window, not a rate.
• OpenAI pushed its Ultrafast tier onto GPT-6.1 Sol, a high-speed mode that is faster and more expensive per token, and retired GPT-5.5 from ChatGPT and Codex. Model retirement forces migration onto whatever the current default is.
• Anthropic cut Sonnet 5.5 cache-read costs by half while its own tooling displayed the old price for two days, a reminder that the published rate and the charged rate can diverge.

Read them together and the pattern is clear. Every provider is simultaneously lowering a headline price and repricing the specific actions heavy users actually perform most. The bundle is the friendliest-looking version of that trade, because the savings are loud and the new meter is quiet.

The practical comparison for a heavy user is not “$30 versus $90.” It is “subscription plus metered third-party usage versus a single flat allowance for all models.” For a developer who lives in one provider’s models, the bundle likely wins. For a researcher or a power user who deliberately spreads work across several frontier models to manage cost and risk, the bundle can win on paper and lose at renewal.

What to do before you switch

1. Audit which models your Cursor workflows actually call. Open your recent sessions and tag each one G (Grok) or T (third-party). If the T share is above roughly 30 percent, run the math with metered pricing before you move.
2. Set a hard spend cap if the tier offers one. A shared pool with no ceiling is the fastest way to an unplanned invoice. If the plan has no built-in cap, treat the subscription as the floor and budget the API bill as a separate line.
3. Keep at least one non-bundled provider active. If all your AI spend flows into a single vendor’s pool, one pricing change reaches your entire stack at once. A second provider on a small independent plan is cheap insurance.
4. Recheck on December 1. Google’s intro pricing expires, and the first full billing cycle of the X Subscription will have landed. That is when real per-request numbers replace launch claims.
5. Watch the November 3 earnings report. SpaceX’s Q3 results will show whether the bundle drove upgrades or merely discounted what subscribers already paid. If tier mix grows, expect the third-party clause to tighten. If it does not, expect more aggressive pricing next.

None of this makes the X Subscription a bad deal. It makes it a conditional one, and the condition is what you actually run. The price went down, the pool went up, and the models you care about most moved to a meter. That is not a contradiction. It is the new standard shape of every AI subscription on the market, and the sooner you price your own workload against it, the less it will surprise you.