Direct answer: OpenAI says it plans to stop supplying models to Cursor after Cursor’s acquisition by SpaceX. November 12, 2026 is the proposed transition date—not a confirmed final deadline—and Cursor could end the current access earlier. Cursor itself is not shutting down.[1][3]
If your work depends on GPT models inside Cursor, do not wait for the final week. Inventory which Cursor features and models you use, choose a replacement path, test it on a non-critical repository, and keep your current setup unchanged until the replacement passes the checks below.
What is changing—and what is not
OpenAI announced on August 28, 2026 that it had notified SpaceX of its intention to wind down the contract that provides OpenAI models to Cursor. OpenAI proposed November 12, 2026 as the shutoff date and said it would not provide future models to Cursor under that agreement.[1]
The date still has an important caveat. OpenAI’s support article calls November 12 a proposed transition date, says Cursor could end access sooner, and says the official termination date will be shared after both companies confirm it.[3]
This is a model-access change, not an announcement that the Cursor editor is closing. Cursor currently documents models from multiple providers, including its own models and models from OpenAI, Anthropic, and Google; availability varies by plan.[4]
The five-minute readiness check
Before changing anything, answer these questions for each person or team using Cursor:
- Which model is selected today? Check the model picker in every workflow that matters.
- Which features are essential? Separate local Chat and Agent from Tab/autocomplete, Auto, Background or Cloud Agents, Automations, CLI, API, and SDK.
- Do you need OpenAI specifically? Record the tasks where changing models produces unacceptable differences.
- How is access paid for? A Cursor subscription, ChatGPT subscription, and OpenAI API account are separate billing paths.
- What is the fallback? Pick a backup model or tool before access changes.
Save the answers in your repository or team runbook. The goal is not to predict what Cursor and OpenAI will negotiate; it is to know which workflows would break if direct OpenAI model access disappears.
Your three documented migration paths
OpenAI currently lists three ways to keep using OpenAI models while working in Cursor: bring your own OpenAI API key for supported local features, run the Codex IDE extension inside Cursor, or use a compatible gateway/provider integration.[3]
Option 1: Bring your own OpenAI API key
This is the closest documented replacement for supported local Chat and Agent requests.
OpenAI’s setup path is:
- Create an OpenAI API key and configure API billing.
- Open Cursor Settings → Models.
- Add and save or verify the OpenAI key.
- Enable an available OpenAI model.
- Select it in local Cursor Chat or Agent.[3]
Cursor’s own documentation similarly says API keys are added under Cursor Settings → Models, after which supported models appear in the picker.[2]
There are three catches:
- API usage is billed through the OpenAI API account; a ChatGPT subscription does not include API usage.[3]
- OpenAI says BYOK applies to supported local Chat and Agent requests, not Cursor Tab/autocomplete, Auto routing, Cloud or Background Agents, Automations, Cursor CLI, or Cursor API/SDK.[3]
- Cursor says its Zero Data Retention policy does not apply when using your own key. Data handling follows the chosen provider’s privacy policy.[2]
Cursor also says the key is sent through encrypted connections to its backend for each request because Cursor performs final prompt building there, but the key is not persisted after the request completes.[2]
Best fit: individuals or teams that need an OpenAI model in local Cursor Chat/Agent and can manage separate API billing and provider-specific data terms.
Option 2: Install the Codex IDE extension in Cursor
OpenAI says the Codex IDE extension can run directly inside Cursor as a separate extension. It can authenticate with an eligible ChatGPT subscription or with an OpenAI API key, and it does not depend on OpenAI models appearing in Cursor’s own model picker.[3]
The documented flow is:
- Install the Codex IDE extension in Cursor.
- Open Codex from the Activity Bar.
- Sign in with ChatGPT or choose API-key authentication.
- Open a project and start the task from the Codex panel.[3]
This does not replace the model behind Cursor Chat, Agent, Tab, Auto, Cloud Agents, or other Cursor services. It gives you a separate Codex agent experience inside the editor.[3]
Best fit: people who want to keep Cursor as the editor while moving OpenAI-powered agent work into a separately authenticated extension.
Option 3: Use a compatible gateway or hosted provider
OpenAI says Cursor can connect through provider-specific settings or a custom OpenAI-compatible base URL. It names Amazon Bedrock and Azure as examples and notes that other compatible gateways may work, but model availability and feature support vary.[3]
The general process is to provision the model with the provider, obtain the endpoint and credentials, configure the provider or custom base URL under Cursor Settings → Models, enable the corresponding model, and test it in a local Chat or Agent session.[3]
Gateway credentials have the same documented feature boundary: they cover supported local Chat and Agent requests, not Tab, Auto, Cloud or Background Agents, Automations, CLI, API, or SDK. OpenAI also warns that custom endpoints may not support every model control Cursor expects.[3]
Best fit: organizations that already centralize AI access, billing, policy, or auditing through Azure, Bedrock, or another gateway.
What not to assume
Do not treat November 12 as final yet
It is the proposed date. OpenAI says it will publish the official termination date after confirmation and that Cursor could end access earlier.[3]
Do not assume a ChatGPT plan pays for API use
OpenAI explicitly separates ChatGPT subscriptions from API billing. A BYOK setup requires an API account with billing configured.[3]
Do not assume BYOK covers every Cursor feature
The documented BYOK scope is supported local Chat and Agent requests. Cursor’s routed and cloud features use models supplied or routed by Cursor.[3]
Do not assume BYOK preserves Cursor’s retention terms
Cursor says its Zero Data Retention policy does not apply to requests made with your own provider key.[2]
Do not expose keys in screenshots or repositories
Paste credentials only into the intended settings field. Do not put API keys in source files, prompts, tickets, recordings, or shared setup documents.
A safe migration test plan
Run this on a small, non-production repository before changing a team-wide default.
1. Capture a baseline
Choose three real tasks you perform often, such as explaining an unfamiliar module, editing code across several files, and diagnosing a failing test. Save the original task text and expected outcome.
2. Configure one replacement path
Add BYOK, the Codex extension, or your approved gateway. Change only one variable at a time so failures have an identifiable cause.
3. Verify identity and billing
Confirm which account is authenticated, which model is selected, and where usage will be billed. If you use Auto, remember that Cursor says the routed model can vary between requests.[4]
4. Test with least privilege
Use a branch without production secrets. Review every proposed command and file change. Do not grant deployment, billing, email, or database permissions merely to test model access.
5. Compare outputs
Check whether the replacement reads the right files, follows repository rules, edits only the requested scope, and produces tests you can execute. Record failures instead of smoothing them over with repeated prompts.
6. Test the feature boundary
If your workflow uses Tab, Auto, Background Agents, Automations, CLI, API, or SDK, test those separately. A successful BYOK Chat request does not prove those features will use your key.[3]
7. Define rollback
Keep the prior configuration available until the replacement passes. Document how to remove the key, disable the extension, restore the previous model, and contact the appropriate account administrator.
Team migration checklist
- [ ] List every repository and team that relies on an OpenAI model in Cursor.
- [ ] Record local versus cloud/routed Cursor features in use.
- [ ] Assign an owner for OpenAI/Cursor status monitoring.
- [ ] Choose BYOK, Codex extension, gateway, another Cursor model, or a mix.
- [ ] Review API billing limits before enabling BYOK.
- [ ] Review provider privacy and retention terms.
- [ ] Test on a non-critical repository.
- [ ] Validate commands, edits, and executable tests.
- [ ] Prepare a fallback model or editor workflow.
- [ ] Recheck the official date before November 12.
- [ ] Remove unused keys and access after migration.
Frequently asked questions
Is Cursor shutting down on November 12, 2026?
No such shutdown was announced. OpenAI announced an intention to end its contract for providing OpenAI models to Cursor; Cursor documents other available model providers and its own models.[1][4]
Is November 12 the confirmed cutoff date?
Not yet. OpenAI calls it a proposed transition date, says Cursor may end access sooner, and says the final date will be shared when confirmed.[3]
Can I keep using OpenAI models in Cursor with my own API key?
For supported local Chat and Agent requests, OpenAI and Cursor both document a BYOK path. It does not cover every Cursor feature, and API usage is billed separately.[2][3]
Will my ChatGPT subscription cover Cursor BYOK usage?
No. OpenAI says ChatGPT subscriptions do not include OpenAI API usage.[3]
Can I use Codex inside Cursor instead?
OpenAI documents a Codex IDE extension that runs as a separate experience inside Cursor and supports eligible ChatGPT or API-key authentication. It does not change the model used by Cursor’s own features.[3]
What should teams do first?
Inventory models and features before choosing a replacement. The critical distinction is whether the workflow uses supported local Chat/Agent requests or Cursor features that cannot use BYOK credentials.
Bottom line
There is time to test, but the access date is not final. Treat November 12 as a planning target, monitor the official OpenAI and Cursor documentation, and move only after your replacement works on real tasks without exposing secrets or changing production systems unexpectedly.[1][3]
Sources
[1] https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex — Our decision on Cursor following its acquisition by SpaceX
[2] https://cursor.com/help/models-and-usage/api-keys — Bring your own API key | Cursor Docs
[3] https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor — Using OpenAI models in Cursor
[4] https://cursor.com/help/models-and-usage/available-models — Available models | Cursor Docs