Will GPT Stop Working in Cursor? What to Check Before OpenAI Model Access Ends on November 12
On August 28, 2026, OpenAI announced that it intends to end its agreement to provide OpenAI models through Cursor, with November 12, 2026 proposed as the shutoff date. The decision follows Cursor’s acquisition by SpaceX.
This is a major change for anyone who currently selects GPT-5.6 Sol, Terra, or Luna in Cursor. However, Cursor itself is not shutting down, and neither are ChatGPT, Codex, or the OpenAI API.
The announcement concerns the model-supply agreement between OpenAI and Cursor. This article separates what is expected to stop on November 12 from what remains undecided, then explains what GPT users can test before the deadline.
Bottom line: Cursor remains, but GPT access inside Cursor is scheduled to end
| Item | What we know as of August 31, 2026 |
|---|---|
| OpenAI models supplied through Cursor | OpenAI plans to end the agreement, with November 12 as the proposed shutoff date |
| Future OpenAI models | OpenAI says they will not be provided to Cursor and specifically mentions its upcoming Astra model |
| Cursor itself | Cursor is not ending; Claude, Gemini, Grok, Composer, and other models remain alternatives |
| ChatGPT, Codex, and the OpenAI API | The announcement does not identify these direct OpenAI services as part of the shutdown |
| BYOK with a personal OpenAI API key | Cursor currently supports some OpenAI models through BYOK, but the announcement does not establish whether that route will remain available after November 12 |
OpenAI describes November 12 as a “proposed shutoff date,” not an unconditional final date. It is sensible to prepare around that date while allowing for further guidance from both OpenAI and Cursor.
What OpenAI announced about ending the agreement
Cursor announced on August 14, 2026 that its acquisition by SpaceX had been completed. Cursor says it plans to use SpaceX’s computing resources to build stronger proprietary models that cost less to run, describing Grok 4.6 as an early example of what the companies can produce together.
Two weeks later, on August 28, OpenAI said it had notified SpaceX of its intention to wind down the agreement that supplies OpenAI models to Cursor. OpenAI is giving the maximum notice period allowed under the agreement and has proposed November 12 as the shutoff date.
According to OpenAI, its custom agreement with Cursor provides a limited cancellation window after a change of control. OpenAI says it cannot be confident that SpaceX will use OpenAI technology within its terms, citing previous contract violations by companies led by Elon Musk.
That is OpenAI’s stated explanation. As of August 31, Cursor’s official blog does not contain a detailed response to the planned model cutoff or a migration plan for users. The competing contractual claims and history cannot be independently resolved from OpenAI’s announcement alone.
The connection to OpenAI models is ending, not Cursor as a whole
Cursor is not an editor built around only one AI model. Its current documentation lists models from OpenAI, Anthropic, Google, SpaceXAI, and other providers, alongside Cursor’s own Composer models.
As of August 31, the model list includes GPT-5.6 Luna, Terra, and Sol; Claude Fable 5, Opus 5, and Sonnet 5; Gemini 3.1 Pro and 3.7 Flash; Grok 4.5 and 4.6; and Composer 2.5.
The Cursor editor and its AI-agent features are therefore expected to remain usable after November 12. Workflows that explicitly pin a GPT model in chats, agents, or automations will need to select a different model.
Rather than thinking that “Cursor will stop working,” a more accurate picture is that one provider connection is being unplugged from an editor that supports several model families.
ChatGPT, Codex, and the OpenAI API are not part of this cutoff
OpenAI’s announcement is specifically about the agreement that supplies its models to Cursor. It does not announce the end of ChatGPT, Codex, or general access to the OpenAI API.
If you want to keep using GPT for coding, Codex is the most direct migration candidate because it is provided by OpenAI. Codex can open a local project, edit files, run commands, and review changes.
Access and billing do not automatically move with the workflow. Usage included in a Cursor plan is separate from a ChatGPT or Codex subscription and from usage billed through the OpenAI API. Before switching, check which OpenAI plan you already have and whether it covers the way you intend to use Codex.
Codex is not a drop-in recreation of Cursor’s interface or workflow. Inline editor completions, Cursor-specific agent features, settings, and pricing will not become identical. Test one real task from beginning to end rather than assuming that changing applications will preserve every part of your workflow.
If you have never used Codex, begin with a small project whose expected result you already understand. That makes it easier to compare the interaction model with Cursor instead of judging only the final code.
Whether a personal OpenAI API key will keep working is still unclear
Cursor supports BYOK, or Bring Your Own Key, which lets you enter an API key from an AI provider you pay for separately. Cursor’s current documentation says personal OpenAI keys can be used with standard, non-reasoning chat models.
However, a personal key does not power features that depend on Cursor’s specialized models, such as Tab Completion. OpenAI’s announcement also does not say whether connections made with personal OpenAI API keys will be permitted after the Cursor supply agreement ends.
BYOK also changes where usage is billed. Cursor says requests made with a personal key go directly to the selected provider at the user’s cost. This can preserve control over the provider account, but it does not recreate every feature or allowance included with a Cursor subscription.
It would therefore be premature to claim that adding an API key will preserve GPT access after November 12. Treat BYOK as a possible option only after Cursor publishes its response and the arrangement is confirmed to comply with OpenAI’s terms.
Auto users may still notice a change
Cursor’s Auto option chooses a model based on the task and current model availability. People who do not manually select a particular GPT model will likely be able to keep using Cursor after November 12 without changing the interface.
That does not mean the experience will be identical. A different model can change the style of the answer, coding preferences, tool behavior, speed, and how quickly included usage is consumed. The label may still say Auto even though the pool behind it no longer contains the same providers.
If Auto handles important work for you, choose a few tasks you request frequently and keep examples of the results. Comparing the same tasks before and after November 12 will make changes easier to identify.
For team use, record the selected model when a result is approved, even if the request began in Auto. Without that history, a later difference can look like a prompt or codebase problem when the underlying model has actually changed.
Five things to do before November 12
- Identify the work that depends on GPT: Record chats, agents, and automations that pin a GPT model, along with tasks where GPT has produced especially reliable results.
- Give the same task to other models: Compare correction accuracy, speed, and usage with Claude, Gemini, Grok, and Composer.
- Run one real project in Codex: Do not stop at a tiny edit. Take a workflow you normally assign to GPT in Cursor and complete the full sequence in Codex.
- Move instructions and rules outside the app: Keep recurring prompts, coding conventions, and procedures in repository Markdown files, Obsidian, or another portable location.
- Check team pricing and permissions: Confirm allowed models, personal API-key policies, on-demand charges, and data-governance requirements with the administrator.
The goal is to avoid switching a model name in a panic on November 11. Models have different strengths and interpret instructions differently, so migration decisions should be based on real tasks.
Choose an alternative based on whether Cursor or GPT matters more
| What you want to preserve | What to test |
|---|---|
| Cursor’s interface, completion, and agents | Compare Claude, Gemini, Grok, and Composer inside Cursor |
| GPT-powered coding | Use Codex directly, or build a suitable workflow with the OpenAI API |
| The specific combination of Cursor and GPT | Wait for additional guidance from Cursor and OpenAI about BYOK |
| No increase in subscription cost | Test whether models already included in your current plan can finish your common tasks |
Do not compare models only by an abstract intelligence ranking. Count revisions, time, and usage until the work is complete. AI Jiten applies the same principle in “Is the Cheapest AI Model Really Cheaper? Choose GPT-5.6 by Cost per Completed Task.”
AI Jiten’s take: do not treat the editor and model as one inseparable product
One of Cursor’s strengths has been the ability to choose models from several companies in one development environment. This contract decision shows that even a capable tool can lose a model because of an acquisition or a change in a relationship between providers.
Under SpaceX, Cursor may increasingly emphasize Grok, Composer, and other models developed within its new corporate group. That direction is not inherently bad. For people who chose Cursor specifically for GPT, however, it changes the product’s role even though the Cursor name and interface remain.
Keep code in Git or a local repository, and do not lock your recurring instructions, rules, and procedures inside a single application. Portable project knowledge makes it easier to switch models within Cursor or move the work to Codex without rebuilding the entire process.
There are roughly two and a half months before November 12, so there is no need to abandon Cursor immediately. Choose one GPT-dependent workflow and test it with both an alternative Cursor model and Codex. Make the migration decision after comparing the results.
Summary
- OpenAI plans to end its model-supply agreement with Cursor and has proposed November 12, 2026 as the shutoff date.
- The decision follows Cursor’s acquisition by SpaceX and relies on a change-of-control cancellation provision.
- Cursor itself is not ending; Claude, Gemini, Grok, Composer, and other models remain available alternatives.
- ChatGPT, Codex, and general OpenAI API access are not identified as part of this cutoff.
- It remains unclear whether BYOK with a personal OpenAI API key will preserve GPT access after November 12.
- Before the cutoff, compare everyday work in alternative Cursor models and Codex, and store instructions in a portable location.
The largest impact will fall on people who explicitly select GPT in Cursor. Check your model settings and usage history first, then give one GPT-dependent task to both another Cursor model and Codex.
Sources (checked August 31, 2026)
- OpenAI: Our decision on Cursor following its acquisition by SpaceX
- Cursor: Cursor is now a part of SpaceX
- Cursor Docs: Models & Pricing
- Cursor Docs: API Keys
The date, affected models, BYOK policy, and Cursor’s available models may change. This article reflects official announcements and documentation from OpenAI and Cursor available as of August 31, 2026.
