The X Subscription folds Grok, Cursor Pro and X Premium+ into $30. If you code in Cursor with Claude, it costs $30 a month more than Cursor Pro, because the Claude usage Pro included is gone.

Short version

• The X Subscription launched October 12 at $0, $8, $30, $100 and $200. Premium, the $30 tier, covers what SuperGrok, Cursor Pro and X Premium+ did for $90 combined.
• Its weekly pool covers Grok models and Composer 2.5 only. Claude and GPT in Cursor bill at API list price from the first token, and only on Premium or higher.
• A Cursor Pro user coding mostly in Claude Sonnet 5.5 pays $30 a month more after switching, at any volume above $20 of usage. Someone who also paid for SuperGrok and X Premium+ saves $40.
• The switch can't be undone, and none of it touches the Grok API.

The five tiers

xAI announced the plan on October 12. It's one bill for Grok, Grok Bot, Cursor and X, and it's what Cursor's pricing page now sells. Pro, Pro+ and the old Ultra carry on as legacy plans, and existing subscribers keep them unless they choose to move.

xAI's announcement and Cursor's usage docs, read October 19, 2026. US prices before tax. No annual billing is listed.

The multipliers are the only thing xAI tells you about pool size. Nobody has published what "base" means in tokens, dollars or requests, so 4x Plus is a ratio with no unit. xAI says each tier gives "at least twice as much" Grok coding usage as the Cursor plan it replaces, and that's the only thing you can check it against.

The pool resets weekly, not monthly. Top-up credit doesn't reset and expires 365 days after purchase. Auto top-up is off unless you turn it on, and you can set a monthly spending cap at grok.com/dashboard.

What a Cursor session costs on each model

Cursor publishes the per-token rate it charges for every model on its models and pricing page, so we can put a number on a session. Ours is a solid afternoon of agent work: 2 million input tokens, 85% of them read from cache, and 30,000 tokens of output.

Our arithmetic: 300K uncached input, 1.7M cache reads, 30K output, at Cursor's listed input / cache read / output rates per million. Composer 2.5 $0.50 / $0.20 / $2.50. Sonnet 5.5 $2 / $0.10 / $10. Grok 4.7 $2 / $0.50 / $6. Opus 5.5 $4 / $0.20 / $20. GPT-5.6 Sol $4 / $0.40 / $20, a promotional rate OpenAI guarantees through November 21 (Cursor doesn't list GPT-6 yet). Cache writes and Cursor's 2x Grok rate above 256K tokens left out.

Grok 4.7 lands 52% above Sonnet 5.5, even though both list $2 input. The difference is the cache read. Grok charges $0.50 per million for a cached token, a quarter of its input price. Sonnet 5.5 charges $0.10. Agent sessions are mostly cache reads, so that one line decides the bill.

Inside the pool that doesn't matter, because you don't pay per token. It matters the week the pool runs out. xAI's announcement says usage beyond the plan bills at API list prices through top-ups, so a Grok session you top up for costs $1.63 where the same work on Sonnet would cost $1.07.

The $20 that went missing

Legacy Cursor plans have two monthly pools, one for Cursor's own models and one called Other Models for Claude, GPT and the rest. Until an August 24 edit, archived copies of Cursor's docs said Pro included at least $20 of that third-party usage. The live page now just says "included usage," but the pool is still there, and Cursor's launch post confirms it keeps running until you switch.

On the X Subscription it's gone. Here's what that does to four kinds of user, at two of our sessions a workday (42 a month):

Our arithmetic. Old plans: Cursor Pro $20 with $20 of third-party usage included and the rest on-demand at the same rates, SuperGrok $30, X Premium+ $40. Row one assumes Grok usage fits in both pools. 42 sessions are $44.94 on Sonnet 5.5 and $89.88 on Opus 5.5.

The middle two rows show the pattern. Once you use more than $20 of Claude a month, Pro's fee effectively came back to you as usage. Premium's $30 comes back as Grok usage you may not want. So the switch costs exactly $30 a month whether you run 25 sessions or 250. Below $20 of Claude it costs less, down to $10 if you barely touch it.

The rows that save money are the ones xAI's marketing points at. If you were already paying for SuperGrok and X Premium+, the bundle pays for the lost $20 twice over. If you code mostly with Grok or Composer, you get a bigger pool for $10 more than Pro and the Grok app thrown in.

Nothing here for API users

Cursor's models and pricing docs say the Grok API is billed separately and doesn't draw from the pool. If you call Grok 4.7 from your own code, you still pay $2 in, $0.50 cached and $6 out per million, doubling at 200K tokens, whatever tier you're on.

The prepaid credit you buy for Claude in Cursor isn't an Anthropic balance either. It only spends inside Cursor, at the same list price you'd pay Anthropic directly, with no batch discount. If you already have an Anthropic or OpenAI key, Claude Code or Codex on your own account is the same tokens without a subscription on top. We priced that in our Claude Code cost per session post.

Before you click switch

Open Cursor's usage dashboard and split last month's spend into Cursor models and everything else. If the everything-else column is above $20 and you don't pay for SuperGrok or X Premium+, stay on Pro. You can't come back: Cursor's help page says the move is permanent, cancels the old plan with a prorated refund, and starts a new billing cycle that day. If you bought Cursor through the App Store, cancel that one yourself, or you'll pay twice.

To see what your Claude or GPT usage costs per token outside Cursor, check the pricing page or run your own numbers in the calculator.

Sources

• xAI: Introducing The X Subscription - tiers, prices, pool multipliers, $90 comparison
• Cursor: models and pricing - per-token rates, pool models, Grok API billed separately
• Cursor: usage and limits - weekly vs monthly pools, legacy Other Models pool
• Cursor: what is The X Subscription? - switching rules, refunds, App Store note
• xAI docs: extra usage and top-ups - prepaid credit, spending caps, auto top-up
• Cursor forum: launch thread - "at least twice as much" Grok coding usage claim