Codex vs Cursor: Which to Use Now That OpenAI Is Leaving Cursor

OpenAI plans to stop supplying its models to Cursor on 12 November 2026. That changes the Codex vs Cursor question. Until now it was a choice of workflow: an agent you hand tasks to, or an editor you code in. After the cutoff, it is also a choice about where you can still use GPT for coding.

The short version: Codex is OpenAI's coding agent, and it comes with ChatGPT plans from $20 a month. Cursor is an AI code editor, now owned by SpaceX and sold through The X Subscription from $8 a month. You can also run Codex inside Cursor, and that route does not depend on the contract that is ending. This guide covers what each tool does, what each costs today, exactly what stops working in Cursor, and which one to pay for on a Mac. Every price was checked against the vendors' own pages on 21 October 2026.

The Key Takeaways

• OpenAI is leaving Cursor: its proposed shutoff date is 12 November 2026, and no newer OpenAI model, GPT-6 Astra included, is coming to Cursor before then.
• Codex is included in ChatGPT plans: Plus at $20 runs Codex with GPT-6.1 Sol on the web, in the terminal, in the IDE extension and on iOS. Pro starts at $100.
• Cursor starts at $8, but not for GPT: Plus runs only Cursor's own models, such as Grok and Composer. GPT needs Premium at $30, billed at API list price on top, until the cutoff.
• Your own API key only partly fills the gap: it covers Cursor's local Chat and Agent, but not Tab, Auto, Cloud Agents, Automations or the Cursor CLI.
• The cleanest route is Codex inside Cursor: OpenAI's Codex extension installs in Cursor and signs in with your ChatGPT plan.

Codex vs Cursor at a Glance

Codex is an agent you delegate coding tasks to, and Cursor is an editor you write code in with AI beside you. Codex runs only OpenAI's models and is paid for through a ChatGPT plan. Cursor runs models from several labs and is paid for through The X Subscription, the bundle SpaceX sells for Grok, Cursor and X. Here is how the two compare as of 21 October 2026.

What Changes on 12 November

On 12 November 2026, under OpenAI's proposal, Cursor loses direct access to OpenAI's models. OpenAI announced the decision on 28 August, two weeks after SpaceX completed its purchase of Cursor.

What OpenAI Announced

OpenAI says it is winding down its contract to supply models to Cursor, and it has given the longest notice that contract allows. Its reason, in OpenAI's announcement, is that it "cannot be confident that SpaceX will use our technology within our terms of service". Until the cutoff, Cursor keeps the OpenAI models it uses today, but OpenAI says it is "not providing future models to Cursor". The post singles out Astra, then OpenAI's upcoming model, as one it has to keep within its terms. Cursor's own model list confirms it: the newest OpenAI models it offers are GPT-5.6 Sol, Terra and Luna, and no GPT-6 model appears at all.

OpenAI also posted the decision on its official X account.

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/

— OpenAI (@OpenAI) August 29, 2026

The date is not final. OpenAI's help article says Cursor "may choose to end access sooner", and that OpenAI will share the official termination date once both companies confirm it. Plan for 12 November and treat anything later as a bonus.

What Still Works in Cursor After the Cutoff

After the cutoff, GPT in Cursor only works through credentials you bring yourself, and only in part of Cursor. OpenAI's guide to using its models in Cursor lists three routes: your own OpenAI API key, the Codex IDE extension, or an AI gateway such as Amazon Bedrock or Azure.

The API key and gateway routes cover local Chat and Agent in the Cursor desktop app. They do not cover Cursor Tab and autocomplete, Auto model selection, Cloud or Background Agents, Automations, the Cursor CLI, or Cursor's API and SDK. Those features keep running on models Cursor supplies or routes itself. An API key is also a separate bill. OpenAI points out that a ChatGPT subscription "does not include OpenAI API usage".

There is a second limit, on Cursor's side. Cursor's help page on API keys describes what an OpenAI key unlocks as "Standard, non-reasoning chat models", and it says the model picker shows which OpenAI models are available. Check that picker before you count on a reasoning model such as GPT-5.6 Sol through your own key.

How Cursor Responded

Cursor co-founder Michael Truell played the impact down about an hour later. In his reply on X, he put OpenAI's share of Cursor traffic at about 5% and said the two companies were talking.

We’re sorry to see that OpenAI put out a note saying they plan to block Cursor users from accessing OpenAI models in three months.

OpenAI models serve about 5% of Cursor user traffic, and we’re speaking with the OpenAI team to resolve this.

Cursor was one of the very first users of OpenAI, we’ve worked closely with their team for years, and we’ve trusted their platform to be neutral infrastructure for our business.

— Michael Truell (@mntruell) August 29, 2026

We found no announcement of a deal as of 21 October. OpenAI's help article still carries the 12 November date, and Cursor's documentation still lists GPT-5.6 as its newest OpenAI model.

Codex vs Cursor: How Each One Works

Codex takes a task and works through it on its own, while Cursor keeps you in the editor writing code. That difference decides how much of the work you still do yourself, and it matters more than any model choice.

Codex: An Agent You Hand Tasks To

Codex is built to be given a job. You describe the change, and Codex inspects the repository, edits files and runs the commands and tests on your machine. On a Mac you can reach it from the desktop app, the terminal, your editor and the web. The ChatGPT desktop app has a Codex mode next to ChatGPT, for running several projects side by side. The Codex CLI runs in the terminal and in scripts, and codex exec runs it non-interactively in CI. The Codex IDE extension works in VS Code, Cursor and Windsurf, and Xcode and JetBrains have their own integrations. For longer jobs, Codex web runs tasks in the cloud and comes back with a result to review. OpenAI recommends GPT-6.1 Sol for complex coding, and says it offers "near-Astra performance" at a lower cost than GPT-6 Astra.

Cursor: An Editor You Code In

Cursor looks and behaves like VS Code, because it is built on the VS Code codebase. You type, and Tab completion predicts the next edit. When you want more, Cursor's Agent edits across files while you watch the diffs. On Premium and above, Cloud Agents run tasks remotely, and Bugbot reviews pull requests. The included usage covers Cursor's own models, Grok 4.7 and Composer 2.5 among them. Claude, GPT and Gemini sit in the same model picker, billed per token. That single picker for models from several labs is the reason most people choose Cursor, and it is the part the OpenAI cutoff shrinks.

Codex vs Cursor Pricing

Codex costs nothing extra if you already pay for ChatGPT, and Cursor's cheapest plan costs less but leaves out every non-Grok model. Here are both price lists, as published on 21 October 2026.

What Codex Costs

Codex is included in ChatGPT plans, so the price is the ChatGPT plan you choose. OpenAI's Codex pricing page lists Plus at $20 a month with Codex "on the web, in the CLI, in the IDE extension, and on iOS". Plus also includes automatic code review and Slack. OpenAI estimates Plus at 15 to 160 local messages every five hours on GPT-6.1 Sol, and 5 to 45 on GPT-6 Astra. Weekly limits may also apply. Pro comes at $100, $200 or $500, and OpenAI says Pro plans "currently have no five-hour limit". When Plus runs out, you can buy ChatGPT credits instead of upgrading. You can also run Codex on an OpenAI API key and pay per token, which suits automation in CI. For which ChatGPT tier fits, see our ChatGPT Plus vs Pro comparison.

What Cursor Costs

Cursor's price depends on which models you want. Cursor's pricing page says of Plus at $8: "Only Grok models are available on this plan". Premium at $30 adds Cloud Agents, Bugbot reviews and third-party models. Super at $100 and Ultra at $200 raise the usage. The included pool refreshes weekly and is shared with the Grok apps. Third-party models like Claude, GPT and Gemini are not part of that pool. They are billed at each model's API list price, from the first request, out of prepaid credit you top up yourself. Until 12 November, GPT-5.6 Sol in Cursor costs $4 per million input tokens and $20 per million output tokens, a rate Cursor labels promotional through 21 November. Existing subscribers on legacy Cursor Pro keep their old plan, with its separate allowance for other models. For the full tier breakdown, see our Cursor pricing guide.

The GPT Price Gap

For GPT-heavy coding, Codex on a flat ChatGPT plan is the cheaper and newer option, even before the cutoff. In Cursor, the newest GPT model is GPT-5.6 Sol, at a promotional $4 and $20 per million tokens on top of a $30 plan. On OpenAI's API, the newer GPT-6.1 Sol lists at $2 and $10 per million tokens, half that rate, and it is not available in Cursor. On ChatGPT Plus, GPT-6.1 Sol comes inside the $20 plan, up to the plan's limits, although OpenAI notes it is on a separate rollout, so check your model picker. Cursor wins on price only if its own models are enough for you, at $8 a month.

How to Run Codex Inside Cursor

You can keep Cursor as your editor and use Codex inside it, signed in with your ChatGPT plan. OpenAI says the Codex extension "does not depend on OpenAI models being available through Cursor's own model picker". That is why its help article points Cursor users to it. The setup takes four steps:

1. Install the Codex IDE extension in Cursor. OpenAI's Codex docs list Cursor next to VS Code and Windsurf as an install target.
2. Open Codex from the Activity Bar.
3. Choose Sign in with ChatGPT to use your ChatGPT plan, or Use API Key to bill an OpenAI API account.
4. Open your project and start a task from the Codex panel.

If you prefer the panel on the right, turn on the Secondary Side Bar under View, Appearance, and drag the Codex view into it. Codex then runs as its own agent. It draws on your ChatGPT plan's limits, not on Cursor's pool or credit. It does not change the model behind Cursor's own Chat, Agent, Tab or Cloud Agents. The cheapest way to have both is Cursor on Free or Plus as the editor, and Codex on ChatGPT Plus doing the GPT work.

Codex and Cursor on a Mac

Both tools run natively on a Mac, and both can work with Xcode. The install paths differ.

Codex comes in two forms on a Mac. In the ChatGPT desktop app for macOS, you choose ChatGPT or Codex in the same window. The Codex CLI installs with curl -fsSL https://chatgpt.com/codex/install.sh | sh, or with Homebrew using brew install --cask codex. Apple added agentic coding to Xcode 26.3 in February, with OpenAI's Codex and Anthropic's Claude Agent built in, so you can use Codex without leaving Xcode. On a Mac, Codex can also operate other apps for you; our guide to Codex controlling your Mac shows how.

Cursor ships as a Mac app with separate downloads for Apple silicon (ARM64), Intel (x64) and a Universal build. Its command-line agent installs with curl https://cursor.com/install -fsS | bash. Cursor reaches Xcode through the MCP bridge that Xcode 26.3 opened to any agent. Our Cursor vs Claude Code comparison walks through that Xcode setup step by step.

Where Claude Code Fits

Claude Code is the third tool in most of these searches, and it works the same way as Codex: an agent included in a chat subscription. Claude Pro at $20 a month, or $17 a month billed yearly, includes Claude Code, and Anthropic's Claude Code extension installs in Cursor just like OpenAI's Codex extension does. The difference is the models. Claude Code runs only Claude, and Codex runs only GPT. Cursor still sells Claude after 12 November, at API list price on Premium and above. Anthropic co-founder Tom Brown said on X that Anthropic will "continue to increase compute to support Claude models in Cursor". If you pay for Claude rather than ChatGPT, the same logic applies with Claude Code in place of Codex. Our Cursor vs Claude Code guide covers that pairing in full.

Codex vs Cursor: Which One Should You Pay For?

Pay for ChatGPT Plus and use Codex if GPT is the model you want to code with. Pay for Cursor if you want an AI editor and are happy with Cursor's own models, or with paying per token for Claude and Gemini.

Choose Codex When

Choose Codex if you want to hand over whole tasks, such as refactors, test suites or dependency upgrades, and review the result rather than every keystroke. It is the only way to code with GPT-6.1 Sol or GPT-6 Astra, since neither ever reached Cursor. It also suits anyone who already pays $20 for ChatGPT Plus, because Codex is already in that plan. And it does not force a new editor on you. Codex runs in the terminal, in the ChatGPT app, in Xcode, and inside Cursor itself.

Choose Cursor When

Choose Cursor if you want to write the code yourself, with Tab completion and an agent working in the same editor. Cursor Plus at $8 is the cheapest way into a full AI editor here, as long as Cursor's own models are enough. Premium earns its $30 if you need Cloud Agents, Bugbot reviews, or Claude and Gemini in one picker. If you are on legacy Cursor Pro at $20 with its allowance for other models, keep it.

When You Want a Second Model, Not a Second Agent

Not every coding question needs an agent with write access to your repository. Sometimes you only want to ask why an error happens, or check GPT's answer against Claude's. You do not have to pick a side for that. Fello AI lets you put the same question to two models in the same window, without a third subscription. It puts GPT, Claude, Gemini, Grok and DeepSeek in one native Mac app.

The Verdict

If GPT is what you code with, Codex has been the better deal since GPT-6 shipped, and from 12 November Cursor's model picker can no longer offer it at all. The fix does not require leaving Cursor. Install the Codex extension, sign in with ChatGPT Plus, and keep Cursor on Plus or Free as the editor. Pay for Cursor Premium only for what Cursor alone does: Cloud Agents, Bugbot, and one picker for Grok, Claude and Gemini. If you never cared which model wrote the code, Cursor Plus at $8 still costs less than any plan with GPT-6.1 Sol, and the cutoff will barely touch you.

Frequently Asked Questions