# Codex vs Cursor: which to use as OpenAI plans to end its model deal Codex vs Cursor: OpenAI's proposed 12 November cutoff and running Codex in Cursor. Cursor Free and Plus cannot use own API keys; new customers need Premium ($30/month) or above, plus API charges. Author: Ravi Parikh Canonical: https://voyager.so/blog/codex-vs-cursor Published: 2026-10-20T10:45:00-07:00 Updated: 2026-10-20T10:45:00-07:00 ![Table titled Codex vs Cursor: who pays for GPT, and which GPT you get, prices as of 20 October 2026. Codex, in the CLI, desktop app, cloud or its extension inside Cursor, has GPT-6.1 Sol, GPT-6 Astra and GPT-6 Luna, paid from your ChatGPT plan's allowance (Plus $20, Pro $100, $200 or $500 a month) then credits, or an API key at $2 input and $10 output per million tokens for GPT-6.1 Sol. Cursor's agent on The X Subscription (Premium $30, Super $100, Ultra $200) has GPT-5.6 Sol, Terra, Luna and older GPT-5 models, paid from prepaid credit at list price, $4 and $20 for GPT-5.6 Sol, with OpenAI's proposed shutoff on 12 November 2026. Legacy Cursor plans (Pro $20, Pro+ $60, Ultra $200) have the same models from the Other Models pool, with the same date. With your own OpenAI API key, Cursor's agent requires Premium ($30/month) or above for new customers, or a legacy Pro plan or above, plus API charges. It uses what your API account and Cursor support, in local Chat and Agent only, billed by OpenAI at API rates. Cursor Free and Plus ($8) support neither own API keys nor OpenAI models in Cursor's agent.](https://cdn.moda.app/voyager/blog/codex-vs-cursor/hero-v2-4433244ffc633565.webp) Codex and Cursor overlap at the agent and differ around it. Cursor is a code editor, a fork of VS Code, with completions as you type, its own agent and a menu of models from several companies. Codex is OpenAI's coding agent, without an editor of its own: it runs in a terminal, in the ChatGPT desktop app, in the cloud and as an extension inside editors, Cursor included, and a ChatGPT plan pays for it. The comparison also shifted this autumn. OpenAI plans to stop supplying its models to Cursor, with 12 November 2026 as its proposed date, and has kept its GPT-6 generation out of the catalog Cursor offers ([OpenAI](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing.md)). The short version: if you want OpenAI's newest models doing the work, use Codex. If you like Cursor's editor, run Codex inside it on your ChatGPT plan, but install the current extension by hand, because the copy in Cursor's store predates GPT-6.1 Sol (details [below](#running-codex-inside-cursor)). If you mostly want Tab completions and inline edits while you write, Claude, Gemini and Grok under one bill, or cloud agents that start from Sentry, PagerDuty, webhooks or a schedule and can drive a browser, choose Cursor ([Tab](https://cursor.com/help/ai-features/tab), [inline edit](https://cursor.com/docs/inline-edit/overview), [models and pricing](https://cursor.com/docs/models-and-pricing), [Automations](https://cursor.com/docs/cloud-agent/automations), [cloud agents](https://cursor.com/docs/cloud-agent)). Plans, prices, model lists and capabilities were read from OpenAI's and Cursor's documentation, pricing pages, changelogs, GitHub and extension registries, and staff posts on Cursor's forum, signed out in the US on 20 October 2026. This is research, not a hands-on run of Cursor, so nothing here measures either tool's code quality. Voyager, the desktop app our team builds, runs Codex on its official runtime with your ChatGPT sign-in, with usage counting against that plan's Codex limits. ## What each is **Codex** runs as an open-source CLI, an IDE extension for VS Code and its forks, the Codex side of the ChatGPT desktop app, cloud tasks at chatgpt.com/codex and the ChatGPT phone app ([openai/codex](https://github.com/openai/codex), [IDE extension](https://learn.chatgpt.com/docs/ide), [Codex Cloud](https://learn.chatgpt.com/docs/cloud)). Signed in with ChatGPT it uses OpenAI's models; with configuration it can also use other providers that speak OpenAI's Responses API, or local models through Ollama or LM Studio ([advanced configuration](https://learn.chatgpt.com/docs/config-file/config-advanced)). Every ChatGPT plan includes it; Plus at $20 a month is the cheapest with every surface ([Codex pricing](https://learn.chatgpt.com/docs/pricing)). **Cursor** is a desktop editor for macOS, Windows and Linux from Anysphere, which SpaceX acquired on 14 August 2026 ([Cursor](https://cursor.com/blog/joining-spacex)). It has Tab completion, inline edits, an agent with Agent, Ask, Plan and Debug modes, an Agents Window for parallel agents, a CLI called `agent`, cloud agents, Bugbot for pull request review and an iOS app ([agent overview](https://cursor.com/docs/agent/overview), [CLI](https://cursor.com/docs/cli/overview), [cloud agents](https://cursor.com/docs/cloud-agent)). Since 12 October 2026 new customers buy it inside The X Subscription, one plan for Cursor, Grok and X; earlier subscribers can keep legacy Cursor plans ([models and pricing](https://cursor.com/docs/models-and-pricing), [legacy plans](https://cursor.com/docs/account/pricing/legacy-cursor-plans)). | Layer | Codex | Cursor | | ---------------------- | ---------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | | Editor | None; a panel in VS Code, Cursor or Windsurf, and JetBrains and Xcode integrations | The product: a VS Code fork with Tab, inline edit and diff review | | Agent | Codex in the CLI, desktop app, IDE extension and cloud | Cursor's agent in the editor, Agents Window, `agent` CLI and cloud | | Models | OpenAI's GPT-6.1 Sol, GPT-6 Astra and GPT-6 Luna on a ChatGPT plan | Cursor's Grok 4.7, 4.6, 4.5 and Composer 2.5, plus Claude, Gemini, GPT-5.6 and others | | Who pays for the model | The ChatGPT plan's allowance, then credits, or an OpenAI API key | Cursor's weekly pool for its own models; third-party models at list price on top | Sources: [Codex models](https://learn.chatgpt.com/docs/models), [IDE extension](https://learn.chatgpt.com/docs/ide), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing); read 20 October 2026. ## Working with each, day to day If you use Cursor's agent rather than its completions, the question is what changes when Codex does that job. Both read your repository, edit files, run commands, use MCP servers, Agent Skills and subagents, and take new instructions while a task is running ([Cursor agent overview](https://cursor.com/docs/agent/overview), [Cursor MCP](https://cursor.com/docs/mcp), [Cursor skills](https://cursor.com/docs/skills), [Cursor subagents](https://cursor.com/docs/subagents), [Codex CLI](https://learn.chatgpt.com/docs/cli), [Codex MCP](https://learn.chatgpt.com/docs/extend/mcp), [Codex skills](https://learn.chatgpt.com/docs/build-skills), [Codex subagents](https://learn.chatgpt.com/docs/agent-configuration/subagents)). The differences are in how you write, review and undo. **Writing code yourself.** Cursor's Tab predicts multi-line and cross-file edits and jumps to the next change; Cmd+K or Ctrl+K rewrites a selection in place ([Tab](https://cursor.com/help/ai-features/tab), [inline edit](https://cursor.com/docs/inline-edit/overview)). Codex has no code completion on OpenAI's feature list for the extension ([IDE extension](https://learn.chatgpt.com/docs/ide)). Its one Tab feature, Composer predictions, suggests your next message to the agent, in the desktop app, for personal Pro accounts, in beta ([Composer predictions](https://learn.chatgpt.com/docs/features/composer-predictions)). **Finding code.** Cursor's agent searches with Instant Grep, an index built on your machine, plus an Explore subagent, and Cursor says it "does not store embeddings of your codebase for search" ([Cursor search](https://cursor.com/docs/agent/tools/search)). Codex explores the repository itself as it works, reading files and running commands in your workspace ([Codex CLI](https://learn.chatgpt.com/docs/cli)). Both read `AGENTS.md` files in the root and subfolders, so project instructions steer either one ([Cursor rules](https://cursor.com/docs/rules), [Codex AGENTS.md](https://learn.chatgpt.com/docs/agent-configuration/agents-md)). **Reviewing and undoing.** Cursor creates checkpoints before significant changes, so you can restore every file to an earlier point in the chat without touching git, and the Agents Window manages diffs, commits and pull requests across agents ([agent overview](https://cursor.com/docs/agent/overview), [Agents Window](https://cursor.com/docs/agent/agents-window)). Codex leans on git: the desktop app's review pane shows unstaged, staged, branch or last-turn changes, takes line comments, and stages, reverts or commits; `/review` runs a separate reviewer over a diff without changing your files ([Codex code review](https://learn.chatgpt.com/docs/code-review)). In the editor extension, each edit appears as a diff you can keep or undo ([IDE extension](https://learn.chatgpt.com/docs/ide)). **Debugging in a running app.** Cursor's Debug mode adds logging, has you reproduce the bug and fixes from the runtime evidence, and its browser tool reads console and network traffic from the editor ([Debug mode](https://cursor.com/docs/agent/debug-mode), [browser](https://cursor.com/docs/agent/tools/browser)). Codex's built-in browser lives in the desktop app; it "isn't available in Codex CLI or the Codex IDE extension" ([Codex browser](https://learn.chatgpt.com/docs/browser)). Two situations show how this plays out, using each tool's documented features rather than a test run: * **A change in a repository you do not know.** In Cursor, the agent searches its local index, you watch edits land in the editor, steer with follow-up messages, and roll back to a checkpoint if a turn goes wrong. In Codex, you start it in the repository with `AGENTS.md` describing the project, steer the running turn, then read the result in the review pane or with `/review` and revert through git; commit before handing it a large task ([Cursor search](https://cursor.com/docs/agent/tools/search), [Cursor agent overview](https://cursor.com/docs/agent/overview), [Codex CLI](https://learn.chatgpt.com/docs/cli), [Codex code review](https://learn.chatgpt.com/docs/code-review)). * **A front-end bug you have to reproduce.** Cursor's Debug mode and browser tool keep the loop inside the editor. With Codex, the desktop app's built-in browser can open your local page; the CLI and the extension inside Cursor have no built-in browser, so you either reproduce the bug yourself or add a browser through an MCP server such as Chrome DevTools ([Debug mode](https://cursor.com/docs/agent/debug-mode), [Codex browser](https://learn.chatgpt.com/docs/browser), [Codex MCP](https://learn.chatgpt.com/docs/extend/mcp)). Choose Cursor if you write much of the code yourself, or want checkpoints and runtime debugging in the editor. Choose Codex if you mostly hand over a task, review it as a git diff, and want OpenAI's newest models doing it. ## OpenAI plans to end its model supply to Cursor On 28 August 2026 OpenAI said it intends "to wind down our contract providing OpenAI models to Cursor, with a proposed shutoff date of November 12, 2026", holding the cancellation to the latest date its contract allows "while not providing future models to Cursor" ([OpenAI](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/)). Its help page adds that "Cursor may choose to end access sooner" and that OpenAI will share the official date once both companies confirm it ([Using OpenAI models in Cursor](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor)). Cursor said the next day that it was talking to OpenAI; neither company had announced a different outcome by 20 October ([Cursor forum](https://forum.cursor.com/t/i-heard-openai-is-about-to-cut-off-gpt-access-for-cursor-is-this-true/169876/6)). As of 20 October: * **Cursor's catalog has no GPT-6.** It lists 17 OpenAI entries, from GPT-5 to GPT-5.6 Sol, Terra and Luna, and no GPT-6.1 Sol, GPT-6 Sol, GPT-6 Astra or GPT-6 Luna ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing.md)). Those are the models OpenAI recommends in Codex ([Codex models](https://learn.chatgpt.com/docs/models)). * **Cursor-supplied GPT-5.6 runs until the cutoff,** billed as a third-party model ([Cursor GPT-5.6 Sol](https://cursor.com/docs/models/gpt-5-6-sol)). * **OpenAI names three ways to keep using its models in Cursor** ([Using OpenAI models in Cursor](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor)): 1. Your own OpenAI API key, for Cursor's local Chat and Agent, billed at API prices. Cursor Free and Plus do not support your own API keys. New customers need Premium ($30/month) or above, plus API charges; legacy Pro, Pro+ and Ultra also support own keys ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing), [bring your own API key](https://cursor.com/docs/settings/api-keys)). 2. The Codex IDE extension, signed in with ChatGPT or an API key. 3. An AI gateway such as Amazon Bedrock or Azure, through Cursor's provider settings or a custom base URL. * **The key and gateway routes are limited.** OpenAI says they do not cover Tab, Auto, cloud agents, Automations, the Cursor CLI or Cursor's API and SDK. Cursor's own key page lists OpenAI support as "Standard, non-reasoning chat models" and bills team plans an extra $0.25 per million tokens on own-key requests ([bring your own API key](https://cursor.com/docs/settings/api-keys)). Neither page says whether GPT-6.1 Sol works this way. If you use Cursor mainly for Claude, Gemini or Grok, little changes. If GPT is why you use Cursor's agent, your options are Codex in or beside Cursor, your own key in Cursor's local agent on an eligible plan, or another model. ## Who pays for GPT in each ### A ChatGPT plan does not pay for Cursor's agent OpenAI lets some apps draw on a ChatGPT plan through Sign in with ChatGPT, with each app's share capped by you ([using your ChatGPT plan in other apps](https://help.openai.com/en/articles/20001542-using-your-chatgpt-plan-in-other-apps-and-sites)). Cursor is not among them ([Sign in with ChatGPT partners](https://learn.chatgpt.com/docs/sign-in-with-chatgpt#partners)); Cursor staff confirmed on 30 September that there is no way to use a ChatGPT subscription in Cursor's agent ([Cursor forum](https://forum.cursor.com/t/sign-in-with-chatgpt/173355/5)). A ChatGPT plan reaches Cursor only through OpenAI's own Codex extension or CLI running inside it. ![OpenAI's Sign in with ChatGPT documentation, Partners section. Apps with ChatGPT plan usage: Amp Code, Conductor, Dactyl, Devin, Hermes Agent, Hyperagent, Kilo Code, Lovable (coming soon), Notion, Vercel, Vorflux and Warp. Apps with sign-in only: Airtable, Canva, GitLab, HubSpot and Supabase. Open-source integrations: OpenClaw, OpenCode, Pi and T3. Cursor is not listed.](https://cdn.moda.app/voyager/blog/codex-vs-cursor/fig-siwc-partners-65a9820e92d62b02.webp) OpenAI's list of apps that can use a ChatGPT plan, read 20 October 2026. Cursor is not on it. ### Plans and what they include As of 20 October 2026, in US dollars a month: | Price | Codex (ChatGPT plans) | Cursor (The X Subscription) | | ---------- | ----------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ | | Free | Desktop app with GPT-6 Luna, as it rolls out | Editor with limited Agent, Chat and Tab on Cursor's models | | $8 | Go: desktop app with GPT-6 Luna | Plus: Cursor's models only, at fixed settings | | $20 or $30 | Plus, $20: every surface, GPT-6.1 Sol, cloud tasks, GitHub review | Premium, $30: 4x Plus usage, cloud agents, Automations, Bugbot, third-party models at list price | | $100 | Pro $100: no five-hour limit | Super: 15x Plus usage | | $200 | Pro $200: a larger allowance | Ultra: 50x Plus usage | Sources: [Codex pricing](https://learn.chatgpt.com/docs/pricing), [About ChatGPT Pro tiers](https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing); read 20 October 2026. Cursor bills $80, $300, $1,000 and $2,000 a year if paid annually; ChatGPT Pro is monthly only, and Pro $500 adds Ultrafast mode. Eligible existing Pro $200 subscribers keep their previous allowance through 29 October 2026, then move to the lower allowance at the same monthly price ([About ChatGPT Pro tiers](https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers)). * **A ChatGPT plan includes GPT usage.** Codex draws on an allowance shared with ChatGPT Work. For Plus, OpenAI estimates 15 to 160 local GPT-6.1 Sol messages per five hours, with weekly limits on top; Pro plans currently have no five-hour limit ([Codex pricing](https://learn.chatgpt.com/docs/pricing)). After that you wait, buy credits or upgrade. * **A Cursor plan includes Cursor's models, not GPT.** The weekly pool, shared with Grok, covers Grok 4.7, 4.6, 4.5 and Composer 2.5. "Third-party models like Claude, GPT, and Gemini aren't part of the included pool", and on Premium and above they cost "the model's API list price from the first request, paid from your prepaid credit" ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing)). Super and Ultra buy more Grok and Composer, not any GPT. * **Legacy Cursor plans still include some GPT.** Pro ($20), Pro+ ($60), Ultra ($200) and Teams keep an Other Models pool that pays for third-party models at list price until it runs out; Cursor no longer publishes its size. Switching to The X Subscription is optional, permanent, and removes the pool ([legacy plans](https://cursor.com/docs/account/pricing/legacy-cursor-plans)). ### What GPT costs per token | Model, and where | Input | Cache write | Cached input | Output | | ----------------------------------------------------- | ----: | ----------: | -----------: | -----: | | GPT-6.1 Sol: Codex with an API key, or the OpenAI API | $2 | $2.50 | $0.10 | $10 | | GPT-6 Luna: same | $0.10 | $0.125 | $0.01 | $0.50 | | GPT-5.6 Sol: Cursor's agent, or the OpenAI API | $4 | $5 | $0.40 | $20 | | GPT-5.6 Luna: same | $0.20 | $0.25 | $0.02 | $1.20 | Per million tokens, standard speed, prompts up to 272K tokens. Sources: [OpenAI API pricing](https://developers.openai.com/api/docs/pricing), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing); read 20 October 2026. Cursor's GPT prices match OpenAI's list prices for the same models. ![Cursor's Models and Pricing documentation, Third-party models section: third-party models like Claude, GPT and Gemini aren't part of the included pool and on Premium and above are billed at the model's API list price from the first request. The price table lists, per million tokens, GPT-5.6 Luna at $0.2 input and $1.2 output, GPT-5.6 Sol at $4 input, $5 cache write, $0.4 cache read and $20 output, and GPT-5.6 Terra at $2 input and $12 output, alongside Claude, Gemini and Muse models.](https://cdn.moda.app/voyager/blog/codex-vs-cursor/fig-cursor-third-party-e1d073b595c323a0.webp) Cursor's third-party price table, read 20 October 2026. The newest OpenAI models listed are the GPT-5.6 family. GPT-5.6 Sol costs twice GPT-6.1 Sol's rate for fresh input and output and four times its rate for cached input, which long agent sessions resend on every turn; in one published Codex measurement, 96% of input tokens were cache hits ([InventiveHQ](https://inventivehq.com/blog/codex-subscription-usage-measured)). For illustration, 2 million cached input tokens, 200,000 fresh and 50,000 output cost $1.10 at GPT-6.1 Sol rates and $2.60 at GPT-5.6 Sol rates, or $1.20 and $2.80 if the fresh tokens are all billed as cache writes. The two agents would not send identical tokens for the same task, so treat that as the price gap, not a forecast. ## Running Codex inside Cursor OpenAI's install page has a Cursor button next to VS Code and Windsurf, and the extension signs in with ChatGPT or an API key ([IDE extension](https://learn.chatgpt.com/docs/ide), [authentication](https://learn.chatgpt.com/docs/auth)). Signed in with ChatGPT, its work counts against your Codex allowance, not Cursor's pool or credit, and it does not change the model behind Cursor's own Chat, Agent or Tab ([Using OpenAI models in Cursor](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor)). ![OpenAI's Codex IDE extension quickstart, Get started in your IDE. Step 1, Install or enable Codex, with buttons for Visual Studio Code, Cursor, Windsurf, Visual Studio Code Insiders, Xcode and JetBrains IDEs. Step 2 says that in VS Code, Cursor or Windsurf you choose the Codex icon or run Codex: Open Codex Sidebar from the Command Palette.](https://cdn.moda.app/voyager/blog/codex-vs-cursor/fig-codex-ide-cursor-646a3f32fc5d857d.webp) OpenAI's own install instructions name Cursor. Read 20 October 2026. **The copy in Cursor's store is old.** Cursor installs extensions from the Open VSX registry through its own marketplace proxy ([Cursor extensions](https://cursor.com/help/customization/extensions)). On 20 October 2026 the newest Codex builds there were from 11 September: 26.908.40401 stable and 26.5908.31748 pre-release. Microsoft's VS Code Marketplace had 26.1007.21434, from 9 October ([Open VSX](https://open-vsx.org/extension/openai/chatgpt), [VS Code Marketplace](https://marketplace.visualstudio.com/items?itemName=openai.chatgpt)). An OpenAI engineer wrote on 1 October: "We have been unable to publish to Open VSX since late July due to a change on their end that has blocked our updates" ([openai/codex issue 32499](https://github.com/openai/codex/issues/32499)). That gap decides which models you see. When GPT-6.1 Sol launched, OpenAI listed version 26.928.31416 of the extension (26.5928.31416 for pre-release) as the minimum for it, a version Cursor's store has not reached ([openai/codex issue 49362](https://github.com/openai/codex/issues/49362)). Three ways around it while the registry is stuck: 1. **Install the current build by hand.** OpenAI's suggested workaround is to download the VSIX for your platform from the VS Code Marketplace and run Extensions: Install from VSIX ([issue 32499](https://github.com/openai/codex/issues/32499)). Uninstall the store copy first, reload the window and start a new Codex chat. Automatic updates are off by default for an extension installed from a VSIX, and Cursor's store cannot supply builds it does not have, so while the gap lasts, repeat the manual install for each new version ([VS Code extension marketplace](https://code.visualstudio.com/docs/configure/extensions/extension-marketplace)). 2. **Run the Codex CLI in Cursor's terminal.** It updates from npm and GitHub, with no registry in between, and signs in with the same ChatGPT account ([Codex CLI](https://learn.chatgpt.com/docs/cli)): ```text npm install -g @openai/codex codex ``` 3. **Use the ChatGPT desktop app beside Cursor** for long tasks, with the editor for reading and fixing the result. Two things change compared with Codex elsewhere: the extension does not support Codex plugins, which need the desktop app or CLI ([Codex plugins](https://learn.chatgpt.com/docs/plugins)), and Cursor's Debug mode and browser tool are not available to the Codex panel. Keep one agent per working tree, or give each its own git worktree, so Cursor's agent and Codex do not overwrite each other. **Cursor or plain VS Code around Codex?** If Codex will do all the agent work, VS Code gets each extension release automatically from Microsoft's marketplace. Cursor earns its place if you also want Tab, Cursor's agent for Claude, Gemini or Grok, or its cloud agents. Cursor Free includes Tab with limited usage ([Cursor pricing and plans](https://cursor.com/help/account-and-billing/pricing)). ## Where else they differ ### Safety defaults Codex starts in an operating-system sandbox: it can write and run commands inside the workspace without asking, commands have no network access, and it asks before writing outside the workspace or reaching the network. The sandbox is enforced by the operating system on macOS, Linux and Windows, including a native Windows sandbox ([approvals and security](https://learn.chatgpt.com/docs/agent-approvals-security), [Windows sandbox](https://learn.chatgpt.com/docs/windows/windows-sandbox)). Cursor's default since version 3.6 is Auto-review: allowlisted calls run, other shell commands run in an OS sandbox on macOS and Linux, and on Windows through WSL2, and the rest go to a classifier model that Cursor says "is not a security boundary". The default network mode allows a built-in list of package-manager and tool domains, and the agent can read files anywhere on the system unless you limit reads to the workspace ([run modes](https://cursor.com/docs/agent/security/run-modes), [agent sandboxing](https://cursor.com/blog/agent-sandboxing)). Choose Codex if you want commands kept off the network by default. With Cursor, limit network and read access before leaving it unattended. ### Cloud agents, schedules and review | | Codex | Cursor | | ------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | Plans | Plus and above | Premium and above; legacy Pro, Pro+ and Ultra | | Machine | Isolated workspace per task in an environment you publish: 2 vCPUs and 8 GiB on Plus, 4 vCPUs and 16 GiB on Pro | Dedicated VM per agent on AWS | | Network | Package managers, custom domains or unrestricted, per environment | On by default; allowlists available | | Charge for the machine | None at launch; model usage counts against the plan | None; model usage at your plan's rates | | Started from | Web, desktop, phone and `codex cloud exec`; `@ChatGPT` in Slack or Microsoft Teams on Enterprise with cloud delegation on; GitHub and Linear mentions run on the legacy cloud | Web, desktop, iOS, Slack, GitHub, Linear and an API | | Browser and desktop use in the cloud | Not supported yet, per OpenAI's list of current limitations | Supported: agents can use desktop and browser tools in their VM | | Runs on a schedule or event | Desktop tasks on a local project (computer on, app running); web tasks on schedules or Gmail, Slack and GitHub events, without a local checkout | Automations: a cloud agent with your repository, on schedules or GitHub, Slack, Linear, Sentry, PagerDuty and webhook events | | Pull request review | `@codex review` and automatic reviews on GitHub, counted as Code Review usage | Bugbot, billed by usage | Sources: [Codex cloud environments](https://learn.chatgpt.com/docs/environments/cloud-environments), [Using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan), [Codex commands](https://learn.chatgpt.com/docs/developer-commands), [Codex scheduled tasks](https://learn.chatgpt.com/docs/automations), [Codex GitHub](https://learn.chatgpt.com/docs/third-party/github), [Codex pricing](https://learn.chatgpt.com/docs/pricing), [Cursor cloud agent security](https://cursor.com/docs/cloud-agent/security), [Cursor cloud agents](https://cursor.com/docs/cloud-agent), [Automations](https://cursor.com/docs/cloud-agent/automations), [Bugbot](https://cursor.com/docs/bugbot), [Cursor staff on cloud agent pricing](https://forum.cursor.com/t/what-is-the-pricing-structure-for-using-cloud-agents/156843/14); read 20 October 2026. Choose Cursor if a team wants cloud agents triggered by Sentry, PagerDuty, webhooks or a schedule, or wants them to check a front end in a browser. Choose Codex if cloud tasks and reviews should come out of a ChatGPT plan you already pay for. ### Data On ChatGPT Plus and Pro, Codex conversations may be used to train OpenAI's models unless you turn training off in data controls; Business, Enterprise and Edu do not train on business data by default ([Using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan)). Cursor's Privacy Mode keeps code out of training by Cursor and its model providers, is on by default for teams and can be enforced by admins ([Cursor privacy](https://cursor.com/help/security-and-privacy/privacy)). Cursor's Zero Data Retention agreement does not cover requests made with your own API key ([bring your own API key](https://cursor.com/docs/settings/api-keys)). ### Moving a setup between them * **Shared as is:** `AGENTS.md` instructions and Agent Skills in `.agents/skills` or `~/.agents/skills` ([Cursor skills](https://cursor.com/docs/skills), [Codex skills](https://learn.chatgpt.com/docs/build-skills)). * **Needs rewriting:** subagents. Cursor looks in `.codex/agents` but expects Markdown with YAML frontmatter; Codex's custom agents are TOML ([Cursor subagents](https://cursor.com/docs/subagents), [Codex subagents](https://learn.chatgpt.com/docs/agent-configuration/subagents)). Cursor's `.cursor/rules/*.mdc` files are not read by Codex. * **Configured separately:** MCP servers and hooks, in `.cursor/mcp.json` and `hooks.json` for Cursor and in `config.toml` and `hooks.json` under `.codex` for Codex ([Cursor MCP](https://cursor.com/docs/mcp), [Codex MCP](https://learn.chatgpt.com/docs/extend/mcp), [Cursor hooks](https://cursor.com/docs/hooks), [Codex hooks](https://learn.chatgpt.com/docs/hooks)). * **One-way import:** Codex's `/import` (CLI) or Settings, then Import (desktop app) brings over a Cursor setup's instructions, settings, skills, MCP servers, hooks, subagents and recent chats, without changing Cursor; the CLI takes up to 50 chats from the last 30 days ([import from another agent](https://learn.chatgpt.com/docs/import)). ## Which to choose Prices as of 20 October 2026. | If your work is mostly | Choose | Because | | ---------------------------------------------------------------- | ----------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Handing tasks to GPT models and reviewing the diff | Codex, from ChatGPT Plus ($20) | GPT-6.1 Sol and GPT-6 Astra are included; Cursor's catalog has neither ([Codex pricing](https://learn.chatgpt.com/docs/pricing)) | | Writing code yourself with completions | Cursor, from Free | Tab and inline edit; Codex lists no code completion ([Tab](https://cursor.com/help/ai-features/tab), [Codex IDE extension](https://learn.chatgpt.com/docs/ide)) | | Cursor's editor with OpenAI's models | Both: Cursor Free or Plus ($0 or $8) plus Codex on a ChatGPT plan | The Codex extension or CLI uses your plan inside Cursor; install the current build ([IDE extension](https://learn.chatgpt.com/docs/ide)) | | Claude, Gemini and Grok under one bill | Cursor Premium ($30) or above | Cursor's models are in the pool; others at list price from one credit balance ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing)) | | Hitting Plus limits most weeks on GPT | Codex on Pro $100, if credits would cost more | Pro plans currently have no five-hour limit ([Codex pricing](https://learn.chatgpt.com/docs/pricing)); a bigger Cursor tier adds Grok and Composer, not GPT ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing)) | | Team agents started by Sentry, PagerDuty, webhooks or a schedule | Cursor Premium seats ($30) or above | Automations run cloud agents on your repository ([Automations](https://cursor.com/docs/cloud-agent/automations)) | | A legacy Cursor plan with GPT in its Other Models pool | Keep it for now; move GPT work to Codex or your own key | OpenAI proposes ending its supply on 12 November, with the date still unconfirmed ([transition guidance](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor)); switching plans would also drop the pool for Claude and Gemini ([legacy plans](https://cursor.com/docs/account/pricing/legacy-cursor-plans)) | Neither company converts its allowances into completed work. OpenAI publishes message ranges for Plus; Cursor publishes multipliers of an unstated base ([Codex pricing](https://learn.chatgpt.com/docs/pricing), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing)). Two independent trackers have priced weekly Codex windows at API list rates, a normalized guide to scale rather than a quota: FinOps LLM found a median of $67 across 27 Plus weeks since August and $175 across six Pro $100 weeks from 27 September to 8 October, with the cheap GPT-6 Luna at 49% of its tokens; InventiveHQ found $331 to $377 in two Pro $100 weeks, resetting 30 September and 3 October, of mostly GPT-6 Sol ([FinOps LLM on Plus](https://finopsllm.com/research/codex-usage-limits-chatgpt-pro), [FinOps LLM on Pro $100](https://finopsllm.com/research/chatgpt-pro-100-real-codex-multiplier), [InventiveHQ](https://inventivehq.com/blog/codex-subscription-usage-measured)). The practical test: if you hit the Plus limit in most weeks and the credits you buy to finish them cost more than $80 in a billing month, Pro $100 is likely the cheaper fix, provided its larger allowance covers your work. Our ChatGPT Pro post goes through the measurements. ## Frequently asked questions **Can I use my ChatGPT Plus or Pro subscription in Cursor?** Only through OpenAI's Codex extension or CLI running inside Cursor. Cursor's own agent takes an OpenAI API key, not a ChatGPT login, and Cursor is not on OpenAI's list of apps that can use a ChatGPT plan ([Sign in with ChatGPT](https://learn.chatgpt.com/docs/sign-in-with-chatgpt#partners), [Cursor forum](https://forum.cursor.com/t/sign-in-with-chatgpt/173355/5)). For your own key in Cursor's agent, new customers need Premium ($30/month) or above, plus API charges; Free and Plus do not support it ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing)). **Is OpenAI removing its models from Cursor?** OpenAI plans to end its contract supplying models to Cursor, with 12 November 2026 as the proposed date, and is not providing new models to Cursor in the meantime ([OpenAI](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/)). It says your own API key, a gateway or the Codex extension can still bring OpenAI models into Cursor, with limits ([Using OpenAI models in Cursor](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor)). **Does Cursor have GPT-6?** Not in the catalog Cursor supplies: it stops at GPT-5.6 Sol, Terra and Luna as of 20 October 2026 ([Cursor models and pricing](https://cursor.com/docs/models-and-pricing.md)). GPT-6.1 Sol, GPT-6 Astra and GPT-6 Luna are in Codex and OpenAI's API ([Codex models](https://learn.chatgpt.com/docs/models), [OpenAI API pricing](https://developers.openai.com/api/docs/pricing)). **Can I use Codex in Cursor?** Yes. OpenAI's IDE extension supports Cursor, and the Codex CLI runs in Cursor's terminal ([IDE extension](https://learn.chatgpt.com/docs/ide)). On 20 October the extension in Cursor's store was a September build, below OpenAI's minimum for GPT-6.1 Sol, so install the VSIX from the VS Code Marketplace or use the CLI ([openai/codex issue 49362](https://github.com/openai/codex/issues/49362)). **Is Cursor cheaper than Codex?** For a new customer who wants GPT, no: ChatGPT Plus at $20 includes GPT-6.1 Sol usage, while Cursor's cheapest current plan with any GPT, Premium at $30, bills it per token on top. Cursor Plus at $8 covers only Cursor's own models, and legacy Cursor plans with an Other Models pool are a different calculation ([Codex pricing](https://learn.chatgpt.com/docs/pricing), [Cursor models and pricing](https://cursor.com/docs/models-and-pricing), [legacy plans](https://cursor.com/docs/account/pricing/legacy-cursor-plans)). **Does Cursor read AGENTS.md?** Yes, in the project root and in subfolders, as Codex does, and both load skills from `.agents/skills` ([Cursor rules](https://cursor.com/docs/rules), [Cursor skills](https://cursor.com/docs/skills), [Codex AGENTS.md](https://learn.chatgpt.com/docs/agent-configuration/agents-md), [Codex skills](https://learn.chatgpt.com/docs/build-skills)). ## Voyager is a desktop AI agent for creative work Describe what you want to make; Voyager plans it, uses your apps and tools, and keeps everything editable. [Sign up](/sign-up)