OpenAI Codex vs Cursor: Which coding agent should you use?
Summary
Derived from sourced facts, not from builder opinions- On the official Terminal-Bench 4.0 board, Codex scored 58.18% with GPT-6 Astra or GPT-6.1 Sol at max effort. Cursor has no entry, so there is no like-for-like test.
- Codex comes with ChatGPT plans: Free, Go $8, Plus $20 and Pro from $100. Cursor has Free, Plus $8 (Grok models only), Premium $30, Super $100 and Ultra $200.
- Cursor bills Claude, Gemini and OpenAI models at API list price outside the plan, and its docs list no GPT-6 models.
- OpenAI proposed ending Cursor's access to its models on November 12, 2026.
Choose OpenAI Codex if
Choose Cursor if
Follow both to get an email when either one changes its price, limits or behavior.
Best by task
Access to GPT-6 models
OpenAI Codex
Choice of model makers
Cursor
Cheapest paid plan
Tied
Plans
OpenAI Codex plans
GPT-6 Luna in the desktop app, rolling out
GPT-6 Luna in the desktop app, rolling out
Web, CLI, IDE and iOS · GPT-6.1 Sol and GPT-6 Luna
Plans at $100, $200 or $500 · no five-hour limit
Per user monthly · $20 billed annually
Cursor plans
Limited usage
Grok models only
4x Plus usage · third-party models at API list price
15x Plus usage
50x Plus usage
From $30 per user
Subscription agents are priced per seat with usage limits, so this compares plans rather than dividing a subscription by a guessed number of tasks.
Side by side
Scores appear side by side only when both were published for the same test. Settings differ between sources, so small gaps can reverse under another harness.
Updates
OpenAI Codex
Switching checklist
- Install Codex with curl -fsSL https://chatgpt.com/codex/install.sh | sh on macOS or Linux, or npm install -g @openai/codex, then run codex and sign in with ChatGPT.
- Turn your .cursor/rules into AGENTS.md. Codex reads AGENTS.md from the repository root down to the folder you work in, and /init writes a starting file.
- If your instructions live in another file, list it in project_doc_fallback_filenames in ~/.codex/config.toml.
- Rewrite each server from Cursor's mcp.json as an [mcp_servers.name] table in ~/.codex/config.toml, or add it with codex mcp add.
- Check the default permissions: Codex can read, edit and run commands in the workspace, and asks before editing outside it or using the network.
- Check the limits for your plan: on Plus, about 15-160 local messages per 5 hours with GPT-6.1 Sol and 5-45 with GPT-6 Astra.
- Run one real task in your own repository with both tools and compare the diffs before moving a team.
FAQ
Is Codex better than Cursor?
Codex scored 58.18% on Terminal-Bench 4.0. Cursor has no result on that board, so there is no direct benchmark comparison.
How much do Codex and Cursor cost?
Codex is included in ChatGPT Free, Go $8, Plus $20 and Pro from $100. Cursor costs Free, Plus $8, Premium $30, Super $100 or Ultra $200.
Can I use OpenAI models in Cursor?
Cursor lists GPT-5.6 models at API list price, but no GPT-6 models. OpenAI proposed ending Cursor's access on November 12, 2026.
What are the Codex usage limits?
On Plus, about 15-160 local messages per 5 hours with GPT-6.1 Sol and 5-45 with GPT-6 Astra. ChatGPT Work shares the same allowance. Pro has no five-hour limit.
Does Codex read Cursor rules?
No. Codex reads AGENTS.md, and other file names only if you list them in project_doc_fallback_filenames. Cursor reads AGENTS.md too.
Who owns Cursor?
SpaceX. Anysphere, Cursor's maker, became a SpaceX company on August 14, 2026.
Can I use Claude models in Codex?
No. Codex uses OpenAI models only.