On this page · 11 sections
- What changed between 28 August and 25 November
- Which OpenAI models are in Cursor today
- Your replacement options
- Price comparison per million tokens
- Decision table: which route fits your team
- Worked cost example: a 25-developer team
- Settings checklist before 25 November
- Billing and contract notes for US and UK buyers
- What we could not confirm
- How eCorpIT can help
- References
Summary. OpenAI told SpaceX on 28 August 2026 that it would wind down the contract that supplies its models to Cursor, proposed a 12 November shutoff and said Cursor would get no future models, naming its upcoming model Astra. On 12 November OpenAI said it had temporarily extended access, and on 16 November a Cursor support engineer wrote in the community forum that access now runs through 25 November, with more updates promised before then. By 21 November neither company had posted a newer date. The OpenAI models still in Cursor are GPT-5.6 Sol at $4 input and $20 output per million tokens, GPT-5.6 Terra at $2 and $12, and GPT-5.6 Luna at $0.20 and $1.20. The closest like-for-like swap is Claude Sonnet 5.5 at $2 and $10, which cuts per-developer usage cost by about 48% in our worked example. Gemini 4 Argon matches that price only until 30 November, because Google's list price doubles on 1 December. Teams that need GPT models can run OpenAI's Codex extension inside Cursor on a ChatGPT plan from $20 a month, rather than rely on own-key routing that neither company has confirmed will keep working.
About this article. For engineering leads, platform teams and finance owners who pay for Cursor and use OpenAI models in it. On 21 November 2026 we checked every date, model, price and plan detail against OpenAI's announcement, Cursor's forum, documentation, pricing page, changelog and blog, OpenAI's API, model, deprecation and Codex pages, Google's Gemini API pricing page and the ECB reference rates. The 25 November date comes from a Cursor staff forum post, not a formal notice, and OpenAI has already moved the date once, so re-check before you act. Vendors change prices often; this is not legal or financial advice, and no vendor paid for inclusion. Researched with AI assistance and checked against the primary sources listed in the references.
What changed between 28 August and 25 November
OpenAI's post says its custom agreement with Cursor gives it a limited window to cancel after a change of control, and that it gave the longest notice the contract allows. Its stated reason is that it cannot be confident SpaceX will use OpenAI technology within its terms of service, given its experience with companies controlled by Elon Musk. It will hold the cancellation as late as it can while supplying no future models, and it names Astra as the upcoming model it has in mind. Cursor had announced on 14 August 2026 that SpaceX had completed its acquisition; that post does not give a price.
Cursor's public response has come through one forum thread rather than its blog or changelog. Its latest changelog entries, Voice Chat with Agents on 6 November and Build Live Apps on 5 November, do not mention OpenAI. (Our governance checklist for Cursor's Build Live Apps release covers the second of those.) The timeline below draws on OpenAI's post, Cursor's blog and forum posts 6, 32, 58 and 60.
| Date (2026) | Event | Source |
|---|---|---|
| 14 August | SpaceX completes its acquisition of Cursor | Cursor blog |
| 28 August | OpenAI tells SpaceX it will wind down the Cursor model contract; proposed shutoff 12 November; no future models | OpenAI post |
| 29 August | Cursor: OpenAI models serve about 5% of Cursor user traffic; Anthropic, Google, Meta and Grok models continue | Cursor forum, post 6 |
| 7 September | Cursor explains that the GPT prompt and tools follow the model name, including for custom models | Cursor forum, post 32 |
| 12 November | OpenAI: access temporarily extended beyond 12 November | OpenAI post |
| 16 November | Cursor: OpenAI has extended access through 25 November; more updates to follow | Cursor forum, post 58 |
| 18 November | Cursor community ambassador: newer OpenAI models will not be added for now | Cursor forum, post 60 |
| 25 November | Current end of announced access | Cursor forum, post 58 |
The 16 November post credits productive ongoing discussions, so a further extension or a narrower deal is possible. Nothing published says what happens after 25 November to teams that call OpenAI through their own API key inside Cursor.
Which OpenAI models are in Cursor today
Cursor's model documentation lists three OpenAI models: GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. Users in the forum thread note that GPT-6, GPT-6.1 and Astra never appeared in Cursor, and on 18 November a Cursor community ambassador said newer models would not be added for now. GPT-5.6 is not on OpenAI's API deprecation list, so the models are not being retired at source; only the Cursor route is in question.
Everything else in Cursor's picker sits outside the OpenAI decision. Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) draw from the weekly usage pool included in paid plans. Third-party models from Anthropic (Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5), Google (the Gemini 4 family, Gemini 3.1 Pro and Gemini 3.8 Flash) and Meta (Muse Wonder) bill at API list price. Cursor's 29 August forum post linked a statement from Anthropic's Tom Brown that Anthropic will keep adding compute for Claude in Cursor.
The model-name rule behind any GPT workaround
Cursor picks its system prompt and editing tools from the model name. A Cursor support engineer explained on 7 September that a custom model whose name contains a built-in ID such as gpt-5.6-sol gets the same GPT prompt and apply_patch editing tools as the built-in entry, even when served through your own API key or base URL. Other names fall back to a generic harness with StrReplace and Write tools. The check is a literal substring, so custom-5.6-sol misses and custom-gpt-5.6-sol-medium matches.
Two consequences follow. A team that points Cursor at OpenAI directly keeps the tuned GPT harness only for the GPT-5.6 family; GPT-6.1 Sol has no built-in ID, so it would run on the generic harness. And a user who relies on that harness has asked whether it will survive once the contract ends, which Cursor has not answered publicly. Do not build a team process on it before Cursor's next notice.
Your replacement options
Most teams will use two of these five routes: a default model inside Cursor for everyday agent work, and a separate path for engineers who need GPT models.
Option 1: switch the default to Claude inside Cursor
Claude Sonnet 5.5 costs $2 input and $10 output per million tokens in Cursor, with cache reads at $0.10. Claude Opus 5.5 costs $4 and $20, the same headline rate as GPT-5.6 Sol, with cache reads at $0.20 against Sol's $0.40. Claude Haiku 5.5 costs $0.10 and $0.50, and its $0.01 cache-read price puts it under the threshold that exempts a model from the Cursor Token Rate, which makes it the nearest cheap substitute for GPT-5.6 Luna. All four Claude models are marked as available with US data residency. Our comparison of Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon for coding covers benchmark and cost-per-task differences.
Option 2: Gemini 4 Argon or Carbon, priced to rise on 1 December
Cursor lists Gemini 4 Argon and Gemini 4 Carbon at $2 input, $0.10 cache read and $10 output, both with US data residency. Those are Google's introductory rates. Google's API pricing page moves both models to $4 input, $0.20 cached and $20 output from 1 December 2026. Cursor bills third-party models at the API list price, so expect its rate to follow, which would put Argon level with Opus 5.5. Gemini 4 Hydrogen, at $1.50 and $7.50 with a $0.075 cache read and no announced change, is the cheaper Google option in Cursor and is exempt from the Cursor Token Rate. Evaluate Argon on December prices, not November ones.
Option 3: Grok 4.7 or Composer 2.5 from the included pool
Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5 draw first from the weekly pool included with each seat, so their marginal cost is zero until the pool runs out. Beyond the pool, Grok 4.7 costs $2 input, $0.50 cache read and $6 output, and Composer 2.5 costs $0.50, $0.20 and $2.50. None of them carries the Cursor Token Rate. This route suits routine edits, tests and refactors; evaluate it on your own repositories before making it the default for complex work.
Option 4: your own OpenAI API key (unconfirmed after 25 November)
Cursor's bring-your-own-key page covers the individual Premium, Super and Ultra plans and the legacy Pro, Pro+ and Ultra plans, as well as Teams and Enterprise; the Plus plan excludes your own keys. For OpenAI keys, the page lists standard, non-reasoning chat models and says the picker shows which are available. On Teams and Enterprise, the $0.25 per million Cursor Token Rate still applies to own-key requests on eligible models. Own-key requests still pass through Cursor's servers for final prompt building, and Cursor's zero data retention policy does not cover them. Most important, neither company has said whether own-key calls to OpenAI from Cursor will keep working after the contract ends; a user raised it in October and got no staff answer. Azure OpenAI is also a listed own-key provider, and nobody has said whether that route is affected either. Treat own keys as a stopgap to test, not a plan.
Option 5: run Codex inside Cursor, or beside it
OpenAI lists Cursor as a supported editor for its Codex IDE extension, alongside VS Code and Windsurf. The extension signs in with a ChatGPT plan or an API key, so GPT access no longer depends on the Cursor contract. Codex is included with ChatGPT Plus at $20 a month, Pro from $100 a month (with $100, $200 and $500 tiers), and Business at $20 per user per month billed annually for two or more users, or $25 billed monthly. Plus includes GPT-6.1 Sol and GPT-6 Luna, so this route also gives engineers the newer models Cursor never received. OpenAI estimates 15 to 160 local GPT-6.1 Sol messages per five-hour window on Plus and standard Business seats, and weekly limits may also apply. For GPT-6.1 API prices, see our GPT-5 and o3 retirement cost comparison.
The trade-off is two agents in one editor. Codex is a separate agent with its own settings, so Cursor rules, skills, subagents and review steps do not carry over to it. A forum user who had built a Codex bridge into Cursor said he still preferred Cursor's own editor and harness, so test both on a pilot group before a full rollout. Avoid community bridges that route a ChatGPT subscription into Cursor's custom-model slot; they sit outside both vendors' supported paths.
Price comparison per million tokens
The figures below come from Cursor's model pricing and GPT-5.6 Sol pages, Google's Gemini API pricing page and OpenAI's API pricing page, checked on 21 November 2026. Cursor prices exclude tax, and on Teams and Enterprise eligible third-party models add the $0.25 per million Cursor Token Rate.
| Model and route | Input | Cache read | Output | Token Rate on Teams | US data residency in Cursor |
|---|---|---|---|---|---|
| GPT-5.6 Sol in Cursor (leaving) | $4.00 | $0.40 | $20.00 | Yes | Yes |
| GPT-5.6 Terra in Cursor (leaving) | $2.00 | $0.20 | $12.00 | Yes | Yes |
| GPT-5.6 Luna in Cursor (leaving) | $0.20 | $0.02 | $1.20 | No (exempt) | Yes |
| Claude Opus 5.5 in Cursor | $4.00 | $0.20 | $20.00 | Yes | Yes |
| Claude Sonnet 5.5 in Cursor | $2.00 | $0.10 | $10.00 | Yes | Yes |
| Claude Haiku 5.5 in Cursor | $0.10 | $0.01 | $0.50 | No (exempt) | Yes |
| Gemini 4 Argon in Cursor to 30 November | $2.00 | $0.10 | $10.00 | Yes | Yes |
| Gemini 4 Argon, Google list price from 1 December | $4.00 | $0.20 | $20.00 | Yes | Yes |
| Gemini 4 Hydrogen in Cursor | $1.50 | $0.075 | $7.50 | No (exempt) | Yes |
| Grok 4.7 beyond weekly pool | $2.00 | $0.50 | $6.00 | No | Yes |
| Composer 2.5 beyond weekly pool | $0.50 | $0.20 | $2.50 | No | Yes |
| GPT-6.1 Sol via OpenAI API or Codex | $2.00 | $0.10 | $10.00 | Not in Cursor | Not in Cursor |
| GPT-6.2 Astra via OpenAI API | $8.00 | $0.40 | $40.00 | Not in Cursor | Not in Cursor |
Cursor adds a 10% uplift for regional data residency on eligible models. GPT-5.6 Sol's Fast mode costs twice the standard rate, and requests above 272,000 input tokens double the input rate and bill output at 1.5 times. OpenAI's promotional GPT-5.6 Sol API price ends on 30 November 2026; OpenAI describes $4 and $20 as cuts of 20% and 33%, which implies $5 and $30 afterwards (our arithmetic). If Cursor's access runs past that date, check whether its Sol rate moves too.
Decision table: which route fits your team
| Choose this | If |
|---|---|
| Claude Sonnet 5.5 as the new default | You used GPT-5.6 Sol or Terra for everyday agent work and want the closest price with US data residency |
| Claude Opus 5.5 | Your GPT-5.6 Sol work was mostly long, hard tasks and you accept roughly the same list price |
| Claude Haiku 5.5 | You relied on GPT-5.6 Luna for cheap, high-volume calls such as summaries or simple edits |
| Gemini 4 Argon or Carbon | Your evaluations favour them and you budget at the 1 December price of $4 and $20 |
| Grok 4.7 or Composer 2.5 | You want to stay inside the included weekly pool and your work is routine edits, tests and refactors |
| Codex extension inside Cursor | Specific engineers need GPT models, including GPT-6.1 Sol, and you can fund a ChatGPT seat for them |
| Own OpenAI key | Only as a short test, and only after Cursor confirms own-key access to OpenAI still works after 25 November |
| Move fully to Codex or another tool | GPT models are central to your workflow and Cursor's remaining models fail your evaluations |
Worked cost example: a 25-developer team
This example is illustrative. The token volumes are assumptions, not measurements, and tokenisers differ by vendor, so replace them with figures from your own Cursor usage analytics.
Assumptions. A 25-developer team on Cursor Teams with Premium seats at $30 per seat per month. Each developer uses 30 million input tokens a month, of which 24 million are cache reads and 6 million are uncached input, plus 1.5 million output tokens. The $0.25 Cursor Token Rate applies to all 31.5 million tokens, because it covers input, output and cached tokens. Cache-write charges are excluded, which slightly understates every option that charges for them. Pound figures use the ECB reference rates for 20 November 2026, crossed to £0.76 per US dollar (rounded to two decimals), before VAT.
Per developer, today on GPT-5.6 Sol: 6M x $4.00 = $24.00; 24M x $0.40 = $9.60; 1.5M x $20.00 = $30.00. Model cost $63.60. Cursor Token Rate: 31.5M x $0.25 = $7.875. Total $71.475.
Per developer on Claude Sonnet 5.5: 6M x $2.00 = $12.00; 24M x $0.10 = $2.40; 1.5M x $10.00 = $15.00. Model cost $29.40, plus the same $7.875. Total $37.275.
Per developer on Claude Opus 5.5: $24.00 + $4.80 + $30.00 = $58.80, plus $7.875. Total $66.675. Gemini 4 Argon gives the Sonnet figure at November prices and the Opus figure at Google's 1 December prices.
| Scenario for 25 developers | Seats | Model spend | Monthly total (USD) | Approx. GBP |
|---|---|---|---|---|
| A. Today: GPT-5.6 Sol in Cursor | $750 | $1,786.88 | $2,536.88 | £1,928 |
| B. Claude Sonnet 5.5 in Cursor | $750 | $931.88 | $1,681.88 | £1,278 |
| C. Claude Opus 5.5 in Cursor | $750 | $1,666.88 | $2,416.88 | £1,837 |
| D. Cursor seats plus ChatGPT Business for Codex, billed monthly | $750 + $625 | Included limits | $1,375.00 | £1,045 |
| E. Grok 4.7 within the weekly pool | $750 | $0 until the pool runs out | $750.00 | £570 |
Scenario B saves $855 a month against today ($2,536.88 less $1,681.88), about 34% of the total bill and about 48% of per-developer usage cost ($34.20 of $71.475). Scenario D assumes agent work moves to Codex and to Cursor's included pool, with no third-party spend in Cursor, and holds only if each developer's GPT work fits inside Business's included Codex limits ($25 x 25 = $625 at the monthly rate); heavy users will need credits or Pro plans. Scenario E is a floor, not a forecast: Cursor states pool sizes as multiples of the Plus plan rather than in tokens, so measure when your team exhausts it. Cursor and OpenAI list prices exclude tax, so UK buyers should add VAT as their finance team advises.
Settings checklist before 25 November
For Cursor admins
- Pull usage by model from the Admin Dashboard, which Cursor's pricing page says shows usage and key metrics, and list every person and team spending on GPT-5.6 Sol, Terra or Luna in the last 30 days.
- Pick a default replacement per use case from the decision table and announce it before 25 November.
- Search shared rules, skills, subagent definitions and team marketplace plugins for hard-coded GPT model names. On legacy plans, a subagent that names a third-party model draws from the third-party pool even when the parent chat runs on Auto, Grok or Composer, so a subagent pinned to GPT may fail or change cost when GPT goes.
- Check Automations, cloud agent setups and any CLI or headless CI scripts for a pinned GPT model name.
- On Enterprise, use the repository, model and MCP access controls to allowlist the replacement models, so developers do not drift to an expensive default.
- Set or lower the monthly additional usage limit, managed under additional usage credits at grok.com/dashboard/billing on The X Subscription, or the on-demand spend limit on legacy plans. Opus 5.5 costs the same as Sol, so a team that swaps blindly will not save money.
- If you run Cursor Router in Balance or Intelligence mode on Teams or Enterprise, review spend after the first week, because routes to third-party models bill as third-party usage.
For developers
- Change the model on saved chats and Projects that default to a GPT model.
- If you test your own key, name the custom model so it contains the built-in ID (for example
team-gpt-5.6-sol) to keep the GPT harness, and record that it may stop working without notice.
- If you need GPT-6.1 Sol, install the Codex extension from inside Cursor and sign in with your company ChatGPT workspace, not a personal account.
- Re-run two or three familiar tasks on the new default and note differences in diffs, test output and follow-through.
Billing and contract notes for US and UK buyers
Cursor's pricing page says subscriptions are sold only through X Subscription Corp, and the plans are bundled with Grok and X benefits under The X Subscription. Paid tiers are Plus at $8 a month (Cursor's own models only), Premium at $30, Super at $100 and Ultra at $200, with annual options of $80, $300, $1,000 and $2,000. Teams seats cost $30, $100 or $200 per user per month, and Enterprise is quoted. Billing for these plans is managed at grok.com rather than cursor.com.
Legacy Cursor plans still run for existing customers: Pro at $20, Pro+ at $60, Ultra at $200, Teams Standard at $40 per user and Teams Premium at $120 per user, plus Start at ₹649 a month in India. The paid legacy plans include an Other Models pool for third-party models, which covers GPT today and will keep covering Claude and Gemini after the cutoff. Cursor describes switching to The X Subscription as optional and permanent, and the switch drops the included third-party allowance, so do not let the OpenAI change push you into a plan move you have not costed.
Ask your Cursor account team in writing whether own-key calls to OpenAI will be blocked, whether the GPT harness will be removed, and whether any credit applies if you bought Cursor partly for OpenAI access. UK buyers should confirm the contracting entity and VAT position after the move to X Subscription Corp. Our roundup of developer service cutoffs due in December 2026 lists other changes landing in the same weeks.
What we could not confirm
- Whether access will end on 25 November or be extended again. The date comes from a forum post, and Cursor says more updates will follow before then.
- Whether own-key access to OpenAI models, directly or through Azure OpenAI, survives the contract's end.
- Whether Cursor will keep its GPT-specific prompt and tools for custom models after the cutoff.
- Whether Cursor's rates for Gemini 4 Argon, Gemini 4 Carbon and GPT-5.6 Sol will change when the vendors' current prices end on 30 November.
- The price SpaceX paid for Cursor, which Cursor's acquisition post does not state.
How eCorpIT can help
eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our AI and machine learning services team can build an evaluation set from your own repositories, compare replacement models on cost and output quality, and update shared rules, subagents and CI scripts so a model switch lands in one pass. Talk to us at /contact-us/.
Last updated: 21 November 2026.