Cursor Without GPT: What to Check Before OpenAI’s Models Leave

OpenAI has proposed 12 November 2026 as the day its models leave Cursor. If you write code in Cursor and GPT is anywhere in your workflow, today is the day to check — not because the editor stops working, but because the model doing some of your work may quietly disappear from the picker. The fix takes about ten minutes, and there are three ways to keep GPT, or a clean switch to another model. Here is the check, in order.

Quick answer: OpenAI's proposed 12 November cutoff would remove GPT-5.6 models from Cursor. Check whether GPT does your work, then pick a replacement or bring your own key.

What actually happens on 12 November?

On 28 August 2026, OpenAI published a post titled “Our decision on Cursor following its acquisition by SpaceX”. In it, OpenAI said it intends to wind down the contract that supplies its models to Cursor, “with a proposed shutoff date of November 12, 2026”, and that it will not supply future models — including its upcoming Astra — to Cursor at all. The trigger is SpaceX’s $60 billion acquisition of Cursor’s parent company, Anysphere, which closed on 14 August 2026. OpenAI’s stated reason, quoted verbatim: “We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts.”

Read the date carefully, because most headlines do not. 12 November is a proposed date, not a confirmed cutoff. OpenAI’s help article, updated in late October, says OpenAI keeps providing the models Cursor uses today until the transition ends, that Cursor may end access sooner, and that OpenAI will share the official termination date once both companies confirm it. Cursor chief executive Michael Truell responded the same night in August, saying OpenAI models serve about 5% of Cursor’s user traffic and that the two companies are still talking to try to resolve the dispute. As of 11 November, neither side had confirmed a final date. Plan as if the date holds — but do not write about it as a done deal, because it is not one yet.

What goes is narrow: the GPT models Cursor itself supplies in its model picker — as of 10 November 2026, GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. What stays is everything else: Claude, Gemini, Grok and Cursor’s own Composer models keep working through their own direct partnerships, and Anthropic has publicly said it will keep adding compute for Claude inside Cursor. Newer OpenAI models such as GPT-6.1 Sol and GPT-6 Astra never reached Cursor’s list, which fits OpenAI’s statement about future models.

Check 1: Is GPT doing your work inside Cursor?

This is the check most people skip, and it is the one that matters. Five per cent of traffic sounds small until you are the person whose agent runs on it. Open Cursor and look at three places. First, the model picker in Chat and Agent: if GPT-5.6 Sol, Terra or Luna is your selected model, that is GPT doing your work. Second, Auto mode: if you leave Cursor on Auto, some of your requests may have been quietly running on GPT, and after the cutoff Auto routes elsewhere — your results can change without you touching a single setting. Third, your usage: the Cursor dashboard shows recent usage by model, so export it and see how much of your last week was actually GPT. If the answer is “none”, you are done; close this tab. If it is “some”, the next check decides what you do about it.

One more thing from the same month, worth thirty seconds while you are here: on 29 October, a tampered build of Anthropic’s Claude Code extension (versions 2.1.312 and 2.1.313) sat in Cursor’s own extension marketplace for about eleven hours and added an attacker’s SSH key to affected machines. We covered the full check in our guide to checking your machine after the tampered Claude Code extension — if you have not run it yet, run it now. It is a separate story from the OpenAI cutoff, but both arrived in the same month, and both reward the same habit: verify what your editor is actually running.

Check 2: Which Cursor features lose GPT?

Not every feature is affected the same way, and this is where people waste money or get surprised. Your own OpenAI API key — the most obvious workaround — only works in some places. The table below is drawn from OpenAI’s help article and Cursor’s documentation, as reported on 10 November 2026.

Cursor feature Own API key or gateway? What to do
Local Chat and Agent Yes Add a key or gateway, or use the Codex extension
Tab and autocomplete No Nothing — Tab uses Cursor’s built-in models
Auto model selection No Pick your GPT model by hand
Cloud or Background Agents, Automations No Switch model, or run GPT tasks in Codex cloud
Cursor CLI No Switch model, or use the Codex CLI
Cursor API and SDK No Switch the model in your code

Two practical consequences follow. First, Tab autocomplete should be unaffected: Cursor’s documentation says Tab uses its built-in models, not the OpenAI contract. Second, if you built agents with Cursor’s SDK, check every place your code names a GPT model — those calls will not be rescued by an API key.

Three ways to keep using GPT in Cursor

1. Bring your own OpenAI API key

This keeps GPT in Cursor’s own Chat and Agent panels, billed to your OpenAI API account. Create a key on the OpenAI API keys page, then in Cursor open Cursor Settings, then Models, find the OpenAI provider, paste the key and click Save. Then turn on the OpenAI model you want and pick it in Chat or Agent. Two catches: a ChatGPT plan does not include API usage, so you need a separately funded API account; and on Teams and Enterprise plans, own-key requests still carry Cursor’s Token Rate of $0.25 per million tokens. Also note the privacy trade-off: Cursor’s Zero Data Retention policy does not cover own-key requests. Cursor says the key travels with each request but is not stored.

2. Use the Codex IDE extension inside Cursor

OpenAI’s Codex extension runs in its own panel and does not depend on Cursor’s model picker. Install it from the Codex IDE extension page, open Codex from the Activity Bar, and sign in with ChatGPT — it then uses your plan’s included Codex usage instead of per-token billing. It also offers newer models than Cursor’s picker ever will: OpenAI’s Codex pricing page listed GPT-6.1 Sol and GPT-6 Luna on Plus as of 10 November 2026. The trade-off is that it leaves Cursor’s own Chat, Agent, Tab and Cloud Agents unchanged — it is a second panel, not a replacement picker. Note too that OpenAI says Plus, Pro and Business plans will share one allowance across ChatGPT and Codex from December, so the “free” feel of the extension has a limit coming.

3. Route OpenAI models through a gateway

If your company already buys OpenAI models through Azure or Amazon Bedrock, connect Cursor to that account. Cursor supports Azure OpenAI and AWS Bedrock keys, Bedrock IAM roles through its dashboard, and a custom OpenAI-compatible base URL; OpenAI says OpenRouter or Vercel AI Gateway may work, depending on the provider. Provision the model, create the credentials and deployment name, enter them in Cursor Settings, then Models, and test in a local Chat or Agent session. Gateway credentials cover only local Chat and Agent, and some model-specific controls may not pass through a custom base URL.

What keeping GPT costs

On individual plans, a Cursor-supplied GPT model and your own key cost about the same per token, because Cursor already bills third-party models at API list price — and own-key requests skip your included usage. As of 10 November 2026, both Cursor’s GPT-5.6 Sol page and OpenAI’s API pricing listed $4 per million input tokens, $0.40 per million cached input tokens and $20 per million output tokens at standard speed for prompts up to 272K tokens. OpenAI says GPT-5.6 Sol’s promotional pricing runs until 30 November 2026. A worked example, using those published rates: one long local Agent session reading 3 million input tokens (2.4 million of them cached) and writing 150,000 output tokens costs about $6.36 on your own key — $2.40 for new input, $0.96 for cached input and $3.00 for output. On Teams or Enterprise, add Cursor’s $0.25 per million Token Rate, about $0.79 on the same session. Long agent sessions are the real cost risk, so measure a week of your own usage before deciding the key is cheap.

Your plan also decides whether any of this applies. As of 10 November 2026, Cursor’s pricing pages showed: Free ($0) has no third-party models, so nothing changes; Plus ($8 a month) offers Grok and Composer only and does not allow your own API keys, so nothing changes either; Premium ($30), Super ($100) and Ultra ($200) can use GPT today and keep it via an own key, the Codex extension or a gateway; Teams and Enterprise (from $30 a seat) have the same options, with the Token Rate still applying to own keys. If you are on Plus, the honest answer is that the cutoff does not touch you — your decision is whether to upgrade or switch models, not how to keep GPT.

Or switch models inside Cursor

For many people, switching is the better move. As of 10 November 2026, Cursor’s models page listed Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5; Gemini 4 Argon, Gemini 3.1 Pro and Gemini 3.8 Flash; Grok 4.7, 4.6 and 4.5; Cursor’s Composer 2.5; and Meta’s Muse Wonder. Do not swap blindly. Take five recent tasks where GPT did well, run them on two candidates, and compare the diffs, the test results and the token cost before committing. Our Claude Sheets guide shows the kind of first-jobs test that works for a new model, and our Gemini 4 Carbon explainer covers the other model family most Cursor users will weigh. If you use ChatGPT’s plan elsewhere, OpenAI’s Sign in with ChatGPT programme also lets Go, Plus and Pro subscribers use their plan in partner tools such as Amp Code, Devin, Kilo Code and Warp — Cursor is not on that list.

Our take: the five-minute audit

The cutoff itself is a small story — about 5% of Cursor traffic, per Cursor’s own chief executive — but the pattern behind it is the real lesson, and it is worth stating plainly. For years, developers picked tools by features: which editor autocompletes better, which agent edits more files. This month showed a new factor: who owns the tool decides which models you are allowed to use. A company you never signed a contract with can remove the model you rely on, and a marketplace you never think about can push code onto your laptop at 2 a.m. Neither problem is unique to Cursor; Cursor just happened to get both in the same month. The durable fix is not a better key — it is a second tool that does not depend on your editor. Keep one terminal agent ready that runs outside Cursor, such as the Claude Code CLI or the Codex CLI, so no single company can stop your workday. And before today ends, do the audit: check your picker, check your Auto setting, pick one path per workflow from the table above, and re-test your saved prompts and agents on the replacement model. Five minutes now beats a broken Thursday morning.

Frequently Asked Questions

Is OpenAI leaving Cursor on 12 November 2026?

12 November 2026 is the shutoff date OpenAI proposed on 28 August 2026, when it said it would wind down its contract supplying models to Cursor. As of 11 November, OpenAI’s help article still calls it a proposed date, talks between the two companies were ongoing, and no final date had been confirmed. OpenAI says it will share the official termination date once both companies confirm it, and that Cursor may end access sooner.

Can I still use GPT in Cursor after 12 November?

Yes, in part. You can add your own OpenAI API key for local Chat and Agent, connect a gateway such as Azure OpenAI or Amazon Bedrock, or run OpenAI’s Codex IDE extension inside Cursor. You lose GPT in the features that only use Cursor-supplied models: Auto, Cloud and Background Agents, Automations, the Cursor CLI, and the API and SDK.

Does my ChatGPT Plus subscription work in Cursor?

Not in Cursor’s model picker. A ChatGPT subscription does not include API usage, so an own key needs a separate API account with billing. Your Plus plan does work in the Codex IDE extension, which you can install in Cursor and sign in to with ChatGPT — it then draws on your plan’s included Codex usage.

Will Cursor Tab or autocomplete stop working?

No. Cursor’s documentation says Tab completion uses Cursor’s built-in models, and custom API keys only work with chat models. The cutoff changes the list of models you can pick for Chat, Agent and Cursor’s other features — not autocomplete.

Why can’t I find GPT-6.1 Sol or GPT-6 Astra in Cursor?

OpenAI said in August 2026 that it would not provide future models to Cursor. As of 10 November 2026, Cursor’s models page listed only GPT-5.6 Sol, Terra and Luna from OpenAI. To use newer OpenAI models beside your Cursor work, install the Codex IDE extension, which offers GPT-6.1 Sol on eligible ChatGPT plans or with an API key.

Sources and methodology

Claims in this guide rest on OpenAI’s own 28 August 2026 post “Our decision on Cursor following its acquisition by SpaceX”, OpenAI’s help article “Using OpenAI models in Cursor” (updated late October 2026), Cursor’s models and pricing documentation, and Cursor chief executive Michael Truell’s public response of 28 August 2026 — as reported and cross-checked on 12 November 2026 against the Evermx case summary of the cutoff, the Apex Logic guide “OpenAI Models in Cursor After November 12: 3 Ways to Keep GPT” (fact-checked 10 November 2026) and the CodeToDeploy checklist “Cursor Users Have 3 Days” (11 November 2026). Prices and plan details are as published on 10 November 2026 and may change; the 12 November date is OpenAI’s proposal, not a confirmed cutoff, and is labelled as such throughout. Desk-researched, not hands-on tested — stated plainly. Verification date: 12 November 2026.

Leave a Reply

Your email address will not be published. Required fields are marked *