For heavy Claude or GPT users, that difference matters more than the burn reports, and the switch can't be reversed.
Summary
On October 20, 2026, Cursor said existing subscribers don't have to move to its unified X Subscription. Current plans, including their third-party model usage, keep working, and no deadline has been set. Switching is permanent, though, and under the unified plan third-party models are billed separately by usage. Cursor is also looking into reports that unified-plan limits run out faster than expected. It links some of the burn to included features such as Grok Imagine and says other cases are still under review.
What happened
Cursor posted the clarification on its community forum. Legacy plans keep working as they do now, third-party model usage included. No migration deadline has been set, and Cursor says it will give advance notice before that changes. Users can still move between legacy tiers such as Pro, Ultra and Pro+. Once someone completes a switch to the unified X Subscription, the legacy plan can't be bought again. Before checkout, Cursor tells users that third-party models are not included in the plan price and are billed separately by usage.
In the same post, Cursor said it is investigating reports that unified-plan usage limits are draining faster than expected. Some of that consumption appears tied to newer included features, including Grok Imagine, and other cases are still under review. The post doesn't say how many accounts are affected, what caused each case, or whether limits, pricing or refunds will change.
Several other stories from the same period are reported but not verified here. Cursor's Agent SDK docs list an npm package under a Grok name. Vercel's changelog says GPT-6.1 Terra is on AI Gateway and that Cursor users can configure the gateway. An open-source project, kindle-ai-usage, shows token totals across coding tools priced at API list rates. According to its own description, it does not reflect actual subscription charges.
What the archive adds
According to TechCrunch's October 13, 2026 coverage in our archive, SpaceXSI bundled X Premium, Grok and Cursor into one subscription. An xAI announcement page (dossier d1) describes the unified plan rolling out gradually. The detail that matters most for costs comes from Cursor's current Models & Pricing page (dossier d3). It says paid plans share a weekly usage pool across Cursor and Grok for Cursor's own models, while third-party models such as Claude, GPT and Gemini are excluded and billed separately. Together with the forum post, this gives a two-part cost structure on the unified plan. Cursor-model usage draws from a pool shared with Grok, and third-party usage is a separate metered bill. Legacy plans, by Cursor's statement, keep their current third-party usage. Our dossier doesn't include the legacy plans' exact third-party allowances, the unified tier prices, or the per-token rates Cursor charges. Readers need to read those off their own billing pages.
Trend signal
AI coding subscriptions are splitting into two parts: a flat-fee pool for the vendor's own models, and pass-through metering for frontier models from other labs. Cursor's unified plan follows that pattern. The fee now buys access to a bundle, and the cost of third-party inference sits with the user. The irreversibility rule makes the trade sharper. A legacy plan works like a grandfathered contract with third-party usage still included, and Cursor has said it won't force anyone off it without notice. Our read is that the decision to switch now comes down to which models you actually use, more than features or branding.
Why now
1) Product bundling. TechCrunch (Oct 13, 2026) and the xAI announcement (d1) report that X Premium, Grok and Cursor are sold as one subscription, which explains why a unified plan exists. 2) Cost separation. Cursor's checkout notice and Models & Pricing page (d3) state that third-party models are excluded from the plan price and billed by usage, shifting frontier-model inference cost onto users. 3) A shared pool. Per d3, Cursor-model usage is pooled with Grok, and Cursor's forum post links part of the unexpected drain to an included feature, Grok Imagine. Hypothesis, not established: making the switch one-way may be meant to wind down the legacy plans that include third-party usage.
Connections to other events
Here is the per-user math, worked out with variables because the dossier doesn't give us prices.
Legacy plan cost = L (your current tier price). That price covers your Cursor-model usage and, per Cursor's statement, your current third-party model usage within whatever limits your tier has.
Unified plan cost = U (tier price) + T, where T is your monthly third-party usage billed separately. Cursor-model usage comes out of a weekly pool that, per d3, is shared with Grok.
1) Mostly Cursor's own models, little Claude or GPT: T is close to zero, so the comparison is roughly U against L plus whatever the bundle adds (Grok, X Premium) that you'd otherwise pay for. The switch can make sense if the pool is big enough. The burn reports are the open risk here, because that pool is exactly where the unexplained drain is happening.
2) Mostly Claude, GPT or Gemini through Cursor: T grows with use and has no ceiling built into the plan price. You'd give up the third-party usage your legacy price already covers and pay for it as you go. Unless U is well below L, our read is that a heavy third-party user is likely to pay more after switching, and can't go back if the numbers come out badly.
3) A mix of both: you're exposed on both sides. Pool drain from causes still under review hits one part of the bill, and metering hits the other.
The burn reports are worth taking seriously, but their cause is unknown. Cursor itself attributes only part of the drain to included features like Grok Imagine. Cost structure is a better reason to wait than the burn bug.
Why it matters
For individuals on legacy plans: the deadline-free policy means waiting costs nothing, and switching can't be undone. Heavy third-party model users have the least reason to move.
For teams: model the unified plan as a fixed fee plus a variable third-party line item. Ask Cursor which included features draw from the shared pool and whether admins can control them.
For builders on Cursor's tooling: the SDK naming and gateway routing are reported, not verified here. Check plan availability and model billing in the docs before you build on either.
What happens next
- Cursor publishes a usage-accounting change for bundled features such as Grok Imagine on unified plans: an exclusion from the pool, a separate allowance, or a toggle. (3 months, 50%). Watch for: A Cursor forum or changelog post naming Grok Imagine or included features alongside a change to how the weekly pool is counted.
- Cursor announces a firm end date for legacy Pro/Ultra/Pro+ plans. (12 months, 40%). Watch for: The advance-notice email or forum post Cursor promised, giving a date after which legacy plans stop renewing.
- Cursor publishes explicit per-token rates for third-party models on the unified plan in its Models & Pricing docs. (6 months, 60%). Watch for: A Models & Pricing page table listing per-model input/output rates for Claude, GPT or Gemini under the X Subscription.
Limits and uncertainty
We don't have the unified tier prices, the per-token rates Cursor charges for third-party models, or the exact third-party allowances on legacy tiers, so the break-even above is structural, not numeric. The cause of the burn is unknown. Cursor names Grok Imagine only for some cases, and the rest could be bugs, heavier agent use, or accounting changes. How many accounts are affected is undisclosed. Cursor may raise limits or change pricing, which would shift the math. The d3 description of the shared pool is a summary of Cursor's docs and may not match the exact accounting. Users who rarely touch third-party models may lose nothing by switching.
Skill to build now
Do a break-even calculation for your own seat. This week, write down your Cursor usage exactly as your Cursor dashboard reports it, split into Cursor's own models and third-party models, plus any third-party spend you already pay. Then compare two totals: (a) your legacy subscription price, which covers that usage, against (b) the unified plan price plus your third-party usage priced at the rates Cursor shows for the unified plan. If (b) is higher, or the rates aren't published yet, don't switch.
Reader takeaway
Before you switch, pull last month's Cursor usage and split it into Cursor-model usage and third-party usage (Claude, GPT, Gemini). If third-party usage is a large share, stay on your legacy plan. You'd be giving up coverage you already pay for in exchange for an uncapped metered bill, and you couldn't come back. Re-check once Cursor publishes the root cause of the burn and the third-party rates.
Open question
Once Cursor finishes its review, will the burn turn out to be a fixable bug, or does sharing the pool between Grok features and coding make the unified plan's allowance inherently hard to predict?