Codex vs Cursor (2026): Which One to Pay For

Pay for Codex if you want OpenAI’s models on a flat ChatGPT plan. Pay for Cursor if you want an editor first, with its own Grok and Composer models in the price. Both have a free tier and an $8 plan.

Codex comes with every ChatGPT plan, from Free to Pro at up to $500 a month. Cursor is now part of SpaceX, and new customers buy it as part of The X Subscription, one subscription for Grok, Cursor and X. Its plans run from free to $200 a month. GPT models are not in Cursor’s price: from the $30 plan up, you pay API list price for them.

Codex now runs inside editors, Cursor included, and Cursor now has a CLI and cloud agents. This guide uses the vendors’ own pages, read on November 3, 2026.

Codex vs Cursor at a glance

Codex vs Cursor pricing

Both are sold inside a bigger subscription. Codex comes with ChatGPT. Cursor comes with The X Subscription, which also covers Grok and X. Here are the rungs side by side.

Codex is in every ChatGPT plan, per OpenAI’s Codex pricing page. Free and Go get GPT-6 Luna in the desktop app, which OpenAI says is subject to rollout. Plus at $20 is where the CLI, the IDE extension, Codex Cloud and the GitHub and Slack integrations come in. Pro has tiers at $100, $200 and $500, and the $500 tier adds Ultrafast mode. Business is $20 a user a month billed yearly for two or more users, or $25 billed monthly.

Cursor sells five individual tiers: Free, Plus, Premium, Super and Ultra. Annual billing is $80, $300, $1,000 and $2,000 a year for the four paid ones, per Cursor’s models and pricing docs. Team seats are $30, $100 or $200, matching Premium, Super and Ultra. Cursor recommends Premium for new AI users, Super for daily agent users and Ultra for power users. Its pricing page also lists X perks, such as a Premium checkmark, which you can skip.

The multipliers do not compare across vendors: Cursor’s are against its own Plus pool, and OpenAI publishes message estimates instead.

Cursor Pro and Pro+ are now legacy plans. They keep working, with a second monthly pool for third-party models. What the change did to Claude pricing is in Claude Code vs Cursor.

Models: what each plan includes

Codex runs OpenAI models. The Plus plan card lists GPT-6.1 Sol and GPT-6 Luna, and OpenAI’s limits table also lists GPT-6 Astra, GPT-6.1 Terra and GPT-6 Sol for Plus. GPT-5.5 retired from Codex with ChatGPT sign-in on October 14, 2026. Fast mode uses included limits 2.5 times faster.

Cursor puts its own models in the plan: Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5. They draw on the weekly pool at different rates. Plus runs them at fixed settings, with Fast off and medium effort. From Premium up, you can pick Claude, GPT and Gemini at API list price, the per-token rate each maker charges developers, paid from prepaid credit from the first request.

The OpenAI models in Cursor’s docs are GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. GPT-5.6 Sol costs $4 per million input tokens and $20 per million output. The GPT-6 models that Codex runs are not on Cursor’s list as of November 3.

For an OpenAI user, that is the main split. In Codex, the newest GPT models are in the subscription. In Cursor, GPT is a metered extra from an older generation.

This guide does not rank the models. Your repo is the test that counts, and both free tiers let you try first.

Usage limits and what happens at the limit

Codex on Plus counts local messages per five-hour window, and OpenAI says “Weekly limits may also apply.” It publishes estimates, not fixed caps. On Plus, GPT-6.1 Sol allows roughly 15 to 160 local messages per five hours, GPT-6 Astra 5 to 45 and GPT-6 Luna 350 to 3,000. OpenAI says “Pro plans currently have no five-hour limit.” Cloud tasks may use more than local messages, and Codex shares its allowance with ChatGPT Work.

When you hit a Codex limit mid-task, the turn in progress can finish. After that, Plus and Pro users can buy credits and keep going, and anyone can run extra local chats with an API key at API rates. Type /status in the CLI, or open the usage page. Our guide to Codex usage limits by plan covers each plan and when the limits reset.

Cursor has one weekly pool for its own models. It is shared across the editor agent, the CLI, cloud agents, Grok Bot and the Grok apps, and unused usage does not roll over. When it runs out, the editor shows a notification. You can wait for the reset, pay from prepaid credit at the same per-model rates, or upgrade. Cursor says requests are never downgraded in quality or speed, per its usage article.

Prepaid credit expires 365 days after purchase, and a monthly limit caps what you spend on extras. GPT in Cursor draws on that credit from the first request, so set the cap first.

Codex can stop you for a few hours, then hand the window back. Cursor’s pool lasts the week, so a heavy Monday can leave little for Friday.

Where each one runs

Codex runs in four places, all on one ChatGPT sign-in.

1. The CLI. Install it with curl, npm or Homebrew, then run codex in a project folder, per the Codex CLI docs. The CLI is open source under Apache-2.0.
2. The IDE extension. It installs in VS Code, Cursor and Windsurf, and Xcode and JetBrains IDEs have their own Codex integrations, per the IDE extension page.
3. The ChatGPT desktop app. On macOS, Windows and Linux, you choose ChatGPT or Codex when you start.
4. Codex Cloud. You point a task at your GitHub repos from the web, the desktop app or your phone. Each task gets its own workspace and keeps working while your computer is asleep. You review the changes and open a pull request.

Cursor is a desktop app for macOS, Windows and Linux. Open it in the Agents window, built for running many agents in parallel, or in the classic editor, which runs VS Code extensions from Open VSX. The CLI installs with one curl line and starts with agent. Cloud agents run in isolated virtual machines and do not need your computer online. Start them from the web, the iOS app, the desktop app, Slack, GitHub, Linear or the API, per the cloud agents docs. They need Premium or above.

How the agents work

Both read files, edit them, run commands and check the result. The differences sit in the defaults.

• Safety: Codex starts with network access off and writes limited to your workspace, enforced by your operating system, per OpenAI’s security docs. In its Auto preset, it asks before it edits outside the folder or runs a command that needs the network. Cursor’s CLI has a /sandbox setting you can turn on or off.
• Planning: Cursor has Agent, Plan and Ask modes, and Shift+Tab switches to Plan, which asks clarifying questions first. In Codex, /permissions switches to read-only when you want to plan without changes.
• Undo: Cursor saves checkpoints before significant changes, stored locally and apart from Git. OpenAI tells you to make Git checkpoints before and after a Codex task.
• Typing help: Cursor’s Tab suggests code as you type, across lines and files. Codex works by task.
• Review: Codex runs /review on local changes, and on Plus it can review pull requests on GitHub when you tag @codex. Cursor’s Bugbot reviews pull requests on Premium and above.
• Parallel work: Cursor’s Agents window can give each task its own Git worktree, and its Projects hand a large job to a coordinator agent. Codex has subagents, worktrees in the desktop app, and cloud tasks that run side by side.

MCP, rules and AGENTS.md

Instructions: Codex reads AGENTS.md before it does any work. It reads ~/.codex/AGENTS.md for every project, then each AGENTS.md from the repo root down to your folder, up to 32 KiB by default, per its AGENTS.md docs. Cursor reads AGENTS.md in the project root and in subfolders, and has its own rules as .mdc files in .cursor/rules, per its rules docs. One AGENTS.md can serve both.

MCP: both connect to MCP servers, so the same server works in either. Codex takes local STDIO servers and Streamable HTTP servers with OAuth sign-in. Add one with codex mcp add, or edit ~/.codex/config.toml. Cursor takes stdio, SSE and Streamable HTTP servers, reads .cursor/mcp.json in a project or ~/.cursor/mcp.json for every project, and has a Marketplace for one-click installs. Its cloud agents can use MCP servers too.

OpenAI notes that every MCP server adds context and uses more of your Codex limit, so turn off the ones you do not need.

Skills, subagents and hooks: both have all three.

Which one should you pay for?

Pick the model first, then the tool.

• You want OpenAI’s newest models on a flat bill: Codex. Start on ChatGPT Plus at $20, and move to Pro when the five-hour window keeps stopping you.
• You want an editor with autocomplete and a model picker: Cursor. Start at Premium ($30), which is what Cursor recommends for new AI users.
• You want the cheapest way to try: both free tiers include an agent. At $8, Cursor Plus gives the editor agent, while ChatGPT Go gives Codex in the desktop app only.
• You hand off tasks and walk away: either works. Codex Cloud comes with Plus at $20. Cursor’s cloud agents start at Premium ($30).
• You want Claude or Gemini too: Cursor, from Premium, at API list price. Codex runs OpenAI models.
• You code all day: compare ChatGPT Pro, which has no five-hour limit for now, with Cursor’s Super or Ultra. Watch how fast your own work drains each.

If GPT is the model you want, pay for ChatGPT. If the editor is the point, pay for Cursor and treat Grok and Composer as your defaults. Weighing Claude Code too? Claude Code vs Codex compares those two.

Using Codex and Cursor together

You can run Codex inside Cursor. OpenAI’s IDE page has a Cursor install link, and OpenAI’s Codex extension is on Open VSX, the registry Cursor installs from, updated October 26, 2026. Sign in with ChatGPT and it runs on your ChatGPT plan, apart from your Cursor pool. The Codex CLI works in Cursor’s terminal too.

That gives you two ways to use GPT inside Cursor, and they bill differently. Cursor’s own agent with a GPT model is metered at API list price from prepaid credit. The Codex extension spends your ChatGPT plan.

One split to try: Cursor for Tab and quick edits on Grok or Composer, Codex for long tasks on GPT-6.1 Sol and for cloud tasks while you are away. Both read AGENTS.md and both take the same MCP servers. At list prices, ChatGPT Plus plus Cursor Plus is $28 a month, or $50 with Premium.

What breaks is the bookkeeping: a Codex window and week, a Cursor weekly pool, and agents running in more than one place.

Use either one with Context Board

Context Board is ours, so weigh the bias. It is a live board you build for yourself, and both tools can report to it.

Codex and Cursor both connect to it over MCP: add the board by its address and sign in on an Allow page. The agent can then post an update when it ships something, say whether it is working, done, failed or needs you, and put a blocking question in Needs you, next to anything else waiting on you.

Codex does more. A small runner on your computer, set up once with one line from Connect, feeds the AI usage widget with what is left of your Codex five-hour window and week, as Codex last heard from OpenAI. The widget draws them as rings with reset times: green at half or more, amber from 20% and red under 20%. It also lists every Codex session on your computer in Agent activity, and each one waiting for your answer or OK in Needs you. Only names, states and percentages leave your computer.

The board does not read Cursor’s own usage or plan. Its agent shows up when it reports over MCP, so ask it to set its status as it starts.

Agents Live Example

AI usageCodexnow

Codex

38%leftCurrent sessionresets in 2h 15m

57%leftThis weekresets Fri 10:00 AM

As Codex last checked, 2m ago

Cursorweb · feature/billing1m

Working

Adding yearly plans to the checkout

Posted: price table updated

Codexapi · main3m

Working

Moving the webhook queue to Postgres

Added the retry tests

Needs you2 waitingnow

• Cursor · billingNeeds you: keep monthly prices as they are?4m
• Codex · apiasks to run the migration9m

Add AI usage and Needs you from the gallery, or start from the Agents template, which has both. Context Board is free for 3 days, then $12 a month.