Codex vs Claude Code
Both are coding agents that read your project, edit files and run commands, and both start at $20 a month. The real differences are which subscription you already pay for, how the usage limits work, and where you want the agent to run.
We have no affiliate deal with OpenAI or Anthropic, and nobody paid to appear on this page.
- already pay for ChatGPT and want coding inside that plan
- want to try it for $0: Free covers quick coding tasks
- code heavily on ChatGPT Pro, which has no 5-hour cap right now
- want per-model limits published in numbers
- already pay for Claude Pro or Max and want one shared limit
- want Opus 5.5 as the default model
- want to build on CLAUDE.md, skills, plugins and hooks
- want annual billing: Pro is $17 a month paid yearly
Codex vs Claude Code at a glance
| What you get | Codex | Claude Code |
|---|---|---|
| Free option | Yes, quick tasks in the desktop app 1 | No; Free plan excludes it 12 |
| Cheapest paid plan | ChatGPT Go, $8, lightweight tasks 1 | Pro, $20, or $17 billed yearly 11 |
| Main plan | Plus, $20 1 | Pro, $20 11 |
| Heavy-use plans | Pro, $100 / $200 / $500 1 | Max $100; Ultra $200 15 24 |
| Models | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Astra, GPT-6 Luna 2 | Opus 5.5 (default), Sonnet 5.5, Haiku 5.5, Fable 5.2 16 |
| Where it runs | ChatGPT desktop and mobile apps, web, CLI, IDE extension, cloud 3 | Terminal, VS Code, Cursor, JetBrains, desktop app, web, mobile 13 |
| Limit style | Published ranges per model, per 5 hours 1 | 5-hour session plus weekly limit, shared with chat 14 |
| Past the limit | ChatGPT credits on Plus and Business 1 | Usage credits at API rates on Pro and Max 17 |
| Project instructions | AGENTS.md 8 | CLAUDE.md; reads AGENTS.md if there is no CLAUDE.md 19 |
| Team seat | Business, $20 yearly or $25 monthly 1 | Team Standard, $20 yearly or $25 monthly 11 |
Which plans include them
Codex comes with every ChatGPT plan, in some form. OpenAI's Codex pricing page lists Free for "quick coding tasks" and Go at $8 for lightweight tasks, both on GPT-6 Luna in the desktop app. Plus at $20 adds GPT-6.1 Sol and opens Codex on the web, CLI, IDE extension and iOS. OpenAI says the newer GPT-6.2 Astra and GPT-6.2 Sol are available on Plus, Pro, Business, Enterprise and Edu. Pro costs $100, $200 or $500 a month, and the $500 tier adds an Ultrafast mode 1.
Claude Code starts at Pro. Anthropic's setup guide says Claude Code needs a Pro, Max, Ultra, Team, Enterprise or Console account, and the free claude.ai plan does not include it 12. Pro is $20 a month, or $17 a month billed yearly; Max is $100 for 5x Pro's usage, and Ultra is $200 for 10x Pro's weekly usage with no five-hour limit on verified accounts 11 15.
Both can also run on a pay-as-you-go API key instead of a subscription. Codex bills API-key use at API rates 1. Claude Code does the same through a Console account, and Anthropic reports an average of about $13 per developer per active day across enterprise deployments 21. For most individuals, the subscription is cheaper and easier to predict.
Tip: if you pay for one of these chat apps already, start with the coding agent it includes. Both companies now fold the agent into the same plan and the same allowance as chat. Our ChatGPT vs Claude comparison covers the rest of each plan.
Usage limits
Codex publishes numbers; Claude Code publishes multipliers. That is the clearest practical difference between the two plans.
On ChatGPT Plus, OpenAI gives a range of local messages per 5-hour window for each model: 17 to 180 on GPT-6.2 Sol, 6 to 50 on GPT-6.2 Astra, 15 to 160 on GPT-6.1 Sol, 5 to 45 on GPT-6 Astra, and 350 to 3,000 on GPT-6 Luna. Local messages and cloud chats share one allowance, and weekly limits may also apply. OpenAI says Pro plans "currently have no five-hour limit" 1.
On Claude, Pro and Max have a session limit that resets every five hours and a weekly limit. The limits are shared across Claude and Claude Code, so a long chat in the morning leaves less for coding in the afternoon. Anthropic does not publish message counts; Max and Ultra are defined as multiples of Pro, and Ultra has no five-hour limit for verified accounts 15 24. If you hit a limit mid-answer, Claude Code shows "Usage limit reached ยท wrapping up" and finishes the response, once per week on Pro 14.
Both let you keep going past the cap. Codex uses ChatGPT credits on Plus and Business 1. Claude uses usage credits billed at standard API rates, which you switch on under Settings, Usage 17. Our guide to Claude usage limits explains how to stretch the Claude allowance.
Terminal, IDE and cloud
Both run everywhere a developer works; the defaults differ. Codex lives inside the ChatGPT apps first. Claude Code started as a terminal tool and now has its own desktop app.
Terminal
Both install with one command. Codex, per OpenAI's CLI page and GitHub README 4 5:
curl -fsSL https://chatgpt.com/codex/install.sh | sh
Claude Code on macOS, Linux or WSL, per Anthropic's setup guide 12:
curl -fsSL https://claude.ai/install.sh | bash
Then type codex or claude in a project folder and sign in with your ChatGPT or Claude account. Codex's CLI code is public on GitHub 5. For Windows and the other methods, see our Claude Code install guide.
IDE
The Codex extension works in VS Code, Cursor and Windsurf, and Xcode and JetBrains have their own Codex integrations 6. Claude Code's extension installs in VS Code and Cursor, and a JetBrains plugin covers IntelliJ, PyCharm and WebStorm 13. Pro and Max cover Claude Code in these IDEs at no extra cost 14.
Cloud
Codex Cloud gives each task its own workspace that "can keep working while your computer is asleep", and you can run several tasks from one environment, then open a pull request 7. Claude Code runs in the browser at claude.ai/code with no local setup, and cloud scheduled tasks keep running when your computer is off 13. Both have a phone app for checking in on running work.
Working on a real codebase
Both read a project instructions file, plan before editing, and review their own diffs. The mechanics are close enough that a team can switch with little rework.
Project instructions
Tie; Claude Code reads both formatsCodex reads AGENTS.md files from your home folder and from each directory down to where you work, up to 32 KiB combined by default 8. Claude Code reads CLAUDE.md, and if a repo has only an AGENTS.md, it reads that instead 19. A repo set up for one agent works with the other.
How much it does without asking
Similar optionsCodex offers "Ask for approval", "Approve for me" (only potentially unsafe actions stop for you) and "Full access" 9. Claude Code has Manual, Accept edits, Plan and Auto modes; in recent versions, Auto, where a classifier blocks risky actions, is the starting mode in the terminal and VS Code 20. Anthropic warns against its bypass mode outside an isolated machine, and OpenAI suggests most people start with "Ask for approval".
Code review
TieCodex's /review command starts a separate reviewer that reads your diff without touching files, and you can comment @codex review on a GitHub pull request or turn on automatic reviews 10. Claude Code ships a /code-review skill that checks the current diff in a fresh subagent, and Anthropic recommends a separate reviewer session so Claude does not grade its own work 22. It also offers automatic review on every GitHub pull request and GitHub Actions for CI 13.
Skills and plugins
Claude Code adds hooksBoth now package repeatable instructions as skills and bundle tools as plugins 23. Claude Code adds hooks, which run a script every time an event happens, such as a linter after each edit 13. See Claude Code skills and plugins for the ones worth installing.
What the output feels like
Depends on the jobWhen we ran an earlier pair head to head in February 2026, GPT-5.3-Codex was about 25% faster, while Opus 4.6 produced the more polished visual and front-end results. Both companies have shipped newer models since, so rerun that kind of test on your own repo before you commit to a plan.
Pricing for teams
Seat prices are identical at the standard tier. Claude adds a Premium seat for heavy users; OpenAI sells Enterprise by quote.
| Tier | Codex (ChatGPT) | Claude Code (Claude) |
|---|---|---|
| Standard seat | Business, $20 yearly or $25 monthly; 2+ users 1 | Team Standard, $20 yearly or $25 monthly 11 |
| Heavy-use seat | Not offered as a seat | Team Premium, $100 yearly or $125 monthly; 5x Standard usage 11 |
| Enterprise | Contact sales 1 | $20 a seat a month, billed yearly, plus usage at API rates; 20 seats minimum 11 |
| Past the seat allowance | ChatGPT credits 1 | Usage credits, set by an admin 18 |
Every Claude Team seat includes Claude Code, and its usage shares one pool with Claude chat and Cowork 18 21. On Codex Business, Plus-level per-model limits apply to Standard seats 1.
How we compared
Plan, price, limit and feature facts come from OpenAI's Codex docs and pricing page and from Anthropic's pricing page, Claude Code docs and help center, all opened on 23 October 2026; each table row links to its page. OpenAI's ChatGPT help center was not reachable for us on that date, so ChatGPT-side facts come from OpenAI's Codex documentation. The note on output quality comes from a head-to-head we published in February 2026. We update this page when either company changes plans or ships a new coding model.
Questions people ask
Is Codex better than Claude Code?
Neither wins on every job. Codex has a free entry point and published per-model limits; Claude Code defaults to Opus 5.5 and has a larger set of skills, plugins and hooks. Most people should use the one bundled with the chat plan they already pay for.
Is Codex free?
Partly. ChatGPT Free and Go include Codex for quick, lightweight tasks on GPT-6 Luna in the desktop app. The CLI, IDE extension, web, GPT-6.1 Sol and the GPT-6.2 models start with Plus at $20 a month.
Can I use Claude Code for free?
No. Claude Code needs Pro, Max, Ultra, Team, Enterprise or a pay-as-you-go Console account. The free Claude plan does not include it.
Which has higher usage limits?
OpenAI publishes ranges, such as 17 to 180 GPT-6.2 Sol messages per 5 hours on Plus, and Pro currently has no 5-hour cap. Anthropic publishes multipliers instead: Max gives 5x Pro's usage and Ultra 10x Pro's weekly usage. Direct comparison is not possible from vendor pages alone.
Can I use both in the same project?
Yes. Keep your rules in AGENTS.md: Codex reads it natively, and Claude Code reads it when the repo has no CLAUDE.md, or you can import it from a CLAUDE.md.
Related
The next model change, in your inbox
195,000 readers get Uncover AI's practical AI briefing. We tell you what changed and whether it changes your pick.
Sources
- Codex pricing (OpenAI)
- Codex models
- Codex documentation home
- Codex CLI
- openai/codex on GitHub
- Codex IDE extension
- Codex Cloud
- Codex AGENTS.md
- Codex permission modes
- Codex code review
- Claude pricing
- Claude Code advanced setup
- Claude Code overview
- Using Claude Code with your Pro or Max plan
- What is the Max plan
- Claude Code model configuration
- Usage credits for paid Claude plans
- Use Claude Code with your Team or Enterprise plan
- How Claude remembers your project (CLAUDE.md)
- Claude Code permission modes
- Claude Code: manage costs effectively
- Claude Code best practices
- Codex skills and plugins
- What is the Ultra plan?