Codex vs Cursor: OpenAI's coding agent against the AI code editor

What is the difference between Codex and Cursor?

Codex is OpenAI's coding agent; Cursor is an AI code editor. With Codex you describe a task, in a terminal, an IDE panel, the ChatGPT desktop app or the cloud, and the agent reads your code, edits files and runs commands. Cursor is a full editor you write code in, with autocomplete, an agent side panel and cloud agents built in.

So this is less a head-to-head than a question of workflow. If you want to hand off tasks and review diffs, Codex fits. If you want AI inside the editor while you type, Cursor fits. The two also combine well: Codex runs as an extension inside Cursor, and both read the same instruction and skill folders.

How does Codex work?

Codex CLI lets you "inspect code, make changes, run commands, and automate repeatable work without leaving your terminal." The openai/codex repo is open source under Apache-2.0. You install it with a standalone script, npm or Homebrew, run codex in a project, and sign in with ChatGPT or another method such as an API key.

The same agent runs in several places:

• Codex CLI, with codex exec for scripts and pipelines.
• Codex IDE extension for VS Code, Cursor, Windsurf and VS Code Insiders, with separate Codex integrations in Xcode and JetBrains IDEs.
• ChatGPT desktop app, ChatGPT on the web and the ChatGPT mobile app, which OpenAI also lists as places to use Codex.
• Codex Cloud, which runs tasks in OpenAI-managed environments.

OpenAI's models page recommends GPT-6.1 Sol for complex coding and agentic work, and GPT-6 Luna for focused, repeatable tasks. See the Codex CLI guide for setup.

How does Cursor work?

Cursor describes itself as "a coding agent for building ambitious software," and it ships as a desktop editor that can import your VS Code settings, keybindings and extensions. Its main pieces:

• Tab, the "AI-powered autocomplete," which suggests multi-line edits, adds imports and jumps to your next edit, even across files.
• Agent, a side panel (Cmd+I) that edits code and runs terminal commands, with checkpoints you can roll back to.
• Cloud Agents, which run in isolated VMs and can be started from the editor, the web, iOS, Slack, Linear, GitHub comments or the API.
• Cursor CLI, a command called agent with a -p print mode for scripts and CI.
• Bugbot, AI code review on pull requests, on Premium and higher plans.

On models, Cursor offers its own (Composer 2.5 and Grok models trained jointly with SpaceXAI) and, on Premium and higher, third-party models from OpenAI, Anthropic, Google and others at API list price.

Codex vs Cursor side by side

How do instructions and skills compare?

AGENTS.md is shared. Codex reads AGENTS.md from ~/.codex/AGENTS.md and then from each folder between the repo root and your working directory; files closer to where you are override earlier guidance. Cursor's rules docs list AGENTS.md as a "simple alternative to .cursor/rules." One file at the repo root reaches both.

Cursor adds .mdc rules. For behavior that only Cursor needs, such as rules that attach to certain file patterns, Cursor uses .mdc files in .cursor/rules/ with description, globs and alwaysApply frontmatter. Codex does not read these. The Cursor rules guide covers the format.

Skills line up neatly. Codex's skills docs say skills "build on the open agent skills standard," and Codex scans .agents/skills from your working directory up to the repo root, plus $HOME/.agents/skills. Cursor's skills docs list .agents/skills/ as a project location and add that Cursor "also loads skills from Claude and Codex directories." So a skill in .agents/skills/ works in both without copying. In Codex you call one with $name or /skills. See the Codex skills guide and the Cursor skills guide.

How do sandboxing and approvals compare?

Codex starts locked down. OpenAI's approvals and security page says that by default "the agent runs with network access turned off," and an OS-enforced sandbox limits it to the current workspace. Sandbox modes are read-only, workspace-write and danger-full-access, combined with an approval policy such as on-request or never.

Cursor uses run modes. It calls Auto-review "the safest useful setup for most people": allowlisted calls run straight away, other shell commands run in a sandbox when they can, and anything else goes to a classifier for review. An Allowlist mode gives you fixed, predictable approvals instead. In Cursor's sandbox, network access for terminal commands is blocked by default apart from a built-in list of common developer domains, and you open more through your settings.

Both end up in a similar place: work inside the project is easy, and anything that leaves it needs a rule or your approval.

How much do Codex and Cursor cost?

Prices below are as of October 2026.

• Codex (pricing) comes with ChatGPT plans. Plus ($20 a month) and Pro (from $100 a month) include Codex on the web, CLI and IDE extension. Business is $20 per user per month billed annually or $25 monthly. Free and Go list Codex in the desktop app only, subject to rollout. With an API key you pay API rates, without the cloud features.
• Cursor (pricing): Free; Plus at $8 a month without third-party models; Premium at $30 a month with 4x Plus usage, Cloud Agents, Bugbot and third-party models from prepaid credit; Super at $100 and Ultra at $200. Teams seats are $30, $100 or $200 a month. Cursor is sold as part of The X Subscription, which also covers Grok and X.

If you want GPT models in Cursor, note that they count as third-party models, billed at API list price on Premium and higher.

Which should you pick?

Codex tends to fit when:

• You already pay for ChatGPT and want the agent included.
• You prefer to delegate whole tasks and review the diff, rather than write code alongside AI.
• You want a sandbox with network off from the first run.
• You want an open source CLI, or local models with --oss.

Cursor tends to fit when:

• You write a lot of code yourself and want fast autocomplete and inline edits.
• You want to switch between models from several providers in one place.
• You want cloud agents you can start from Slack, Linear or your phone, and PR review with Bugbot.
• Your team already standardizes on a VS Code style editor.

Neither list is a verdict, and you do not have to choose. For a broader view, compare Codex vs Claude Code and Claude Code vs Cursor.

Can you use Codex and Cursor together?

Yes, and the setup is short:

1. Install the Codex extension in Cursor from OpenAI's IDE extension page, or run codex in Cursor's terminal.
2. Keep shared project instructions in AGENTS.md at the repo root. Both tools read it.
3. Keep shared skills in .agents/skills/. Both tools load them.
4. Keep Cursor-only rules in .cursor/rules/.

If you are moving from Cursor to Codex instead, run /import in the Codex CLI and choose Cursor to bring over supported setup, project files and recent chats.

Keep your skills in one place

The skills and prompts you build up are worth more than either tool. Kipt keeps them as Markdown in a private GitHub repo you own and installs every skill into Codex or Claude Code with one npx skills add command. The free SKILL.md generator turns a prompt you keep reusing into a skill folder in your browser.