Is Codex better than Claude Code?

Neither wins every task for me. Codex is faster, lighter and more hands on, and it can miss the overall goal. Claude Code takes a lot of context and understands what is important, and it feels slower. On smaller projects I tend to use Claude Code.

Why did you cancel Codex?

I am already paying for other AI tools alongside Claude Code, and the total is quite a lot. If I were paying for less I would pay for Codex. If I were paying a bit more I would pay for Claude Code.

Codex vs Claude Code price: what does each cost?

Codex is included in ChatGPT Free, Go ($8 a month), Plus ($20 a month) and Pro (from $100 a month): https://developers.openai.com/codex/pricing. Claude Code is included in Claude Pro ($20 a month billed monthly), Max ($100 a month) and Ultra ($200 a month): https://claude.com/pricing. Both pages read 4 November 2026.

Can Claude Code read AGENTS.md?

Yes, since v2.1.277. By default it reads AGENTS.md only when there is no CLAUDE.md in your working directory or above it. If you keep a CLAUDE.md, import the file with @AGENTS.md: https://code.claude.com/docs/en/memory

Does Codex read AGENTS.md?

Yes. Codex reads AGENTS.md files before doing any work, layering global guidance with project-specific overrides: https://developers.openai.com/codex/guides/agents-md

What is the American engineer and German engineer comparison?

Peter Steinberger, who created OpenClaw, told Lex Fridman that Opus was a little bit too American, and Lex answered that Codex is German. I have heard it applied to Claude Code and Codex: https://lexfridman.com/peter-steinberger-transcript

How should you compare Codex and Claude Code?

Use a greenfield project. Add skills and an AGENTS.md that match the project, and run both tools on it. A project where you already use one tool deeply makes the comparison hard.

What changes for Codex usage in December?

OpenAI says Plus, Pro and Business plans share one usage allowance across ChatGPT and Codex from December, measured in tokens: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits

Codex vs Claude Code: I Cancelled Codex but Still Say Try Both

I just cancelled Codex.

I still tell people to try both.

I have used Codex at work and on a side project, next to Claude Code.

This post is the Codex vs Claude Code comparison from that use.

It is my judgement from daily work.

I did not run a measured test, so there are no timings or costs per task here.

Codex vs Claude Code: where each one pulls ahead for me

Codex is fast and hands on.

Claude Code understands what matters in a pile of context.

What I notice	Codex	Claude Code
Speed	Super fast, more hands on, and light	Feels slow on one task
Questioning the task	At times it does not question	These days it can be afraid to write anything off as not needed
The overall goal	At times it misses the point	I trust it with a lot of context to find what is important
My default on smaller projects	No	Yes

Codex is fast and light, and I used it a lot for code review

I tried Codex with Matt Pocock's wayfinder skill.

Wayfinder plans a big piece of work, more than one agent session can hold, as a shared map of decision tickets on your issue tracker.

Codex ran it much faster, and it felt much lighter.

I still use the Claude Code version of wayfinder a bit more.

I also used Codex a lot for code reviews.

At first it was comment reviews. Then I used it for critique reviews too.

Different agents and models are good at reviewing, so a second tool earns its place there.

In general I find Codex good and super fast.

It does not feel as clever to me as Claude Code.

At times it does not question the task.

Both tools load a skill from a SKILL.md file, so a skill like wayfinder runs in either. The docs are on the Codex skills page and the Claude Code skills page.

The American engineer and the German engineer

I have heard the two tools described as an American engineer and a German engineer.

The line goes back to Peter Steinberger, who created OpenClaw, on the Lex Fridman podcast.

He said Opus was "a little bit too American".

Lex Fridman answered that this was because Codex is German.

Steinberger was comparing the models, Opus and Codex.

I have heard it applied to the two tools.

I used to find it true.

At first Codex feels super clean and lightweight.

Then at times it misses the point and the overall goal of the work.

I saw it while building an MVP.

Codex was super slow to understand what was key and what to drop.

It felt over-engineered for an MVP, where being a bit scrappy is better.

Claude Code gets the goal, and sometimes keeps too much

With Claude Code I am confident giving it a lot of context.

It understands what is important in that context.

These days it can also feel afraid to write anything off as not needed.

Claude Code feels slow, and running agents in parallel hides it

Claude Code feels super slow to me on a single task.

Run in parallel, the slowness stops mattering.

Parallel work has its own weaknesses.

When I ran fifteen agents at once, review took longer than everything before it.

Codex feels fast and more hands on.

That has its weaknesses too.

I do not think it is one or the other.

Which should you pick

If I were paying for less, I would pay for Codex.

If I were paying a bit more, I would pay for Claude Code.

I am already paying for other AI tools alongside Claude Code, and the total is quite a lot.

That is why I cancelled Codex.

I still recommend trying both.

The best place to try them is a greenfield project.

Start adding skills and an AGENTS.md that match the project.

It is hard to compare the two on a project where you have already used one tool deeply.

Codex reads AGENTS.md before it does any work.

Claude Code can too. Since v2.1.277 it reads AGENTS.md when there is no CLAUDE.md in your working directory or above it, per the Claude Code memory docs.

If you keep a CLAUDE.md, import the file with @AGENTS.md.

One AGENTS.md then serves both tools. The details are in CLAUDE.md vs AGENTS.md.

If you start with Codex, the Codex config.toml post covers where its settings live.

What Codex and Claude Code cost

Both come with a paid individual plan at $20 a month.

Codex is included in ChatGPT Free, Go ($8 a month), Plus ($20 a month) and Pro (from $100 a month), per the Codex pricing page.

Claude Code is included in Claude Pro ($20 a month billed monthly, $17 a month billed annually), Max ($100 a month) and Ultra ($200 a month), per Anthropic's pricing page.

I read both pages on 4 November 2026.

OpenAI's help center says that from December, usage on Plus, Pro and Business plans is shared across ChatGPT and Codex, and measured in tokens.

Check that page if you rely on Codex alongside ChatGPT.

Frequently asked questions

Is Codex better than Claude Code?: Neither wins every task for me. Codex is faster, lighter and more hands on, and it can miss the overall goal. Claude Code takes a lot of context and understands what is important, and it feels slower. On smaller projects I tend to use Claude Code.
Why did you cancel Codex?: I am already paying for other AI tools alongside Claude Code, and the total is quite a lot. If I were paying for less I would pay for Codex. If I were paying a bit more I would pay for Claude Code.
Codex vs Claude Code price: what does each cost?: Codex is included in ChatGPT Free, Go ($8 a month), Plus ($20 a month) and Pro (from $100 a month): https://developers.openai.com/codex/pricing. Claude Code is included in Claude Pro ($20 a month billed monthly), Max ($100 a month) and Ultra ($200 a month): https://claude.com/pricing. Both pages read 4 November 2026.
Can Claude Code read AGENTS.md?: Yes, since v2.1.277. By default it reads AGENTS.md only when there is no CLAUDE.md in your working directory or above it. If you keep a CLAUDE.md, import the file with @AGENTS.md: https://code.claude.com/docs/en/memory
Does Codex read AGENTS.md?: Yes. Codex reads AGENTS.md files before doing any work, layering global guidance with project-specific overrides: https://developers.openai.com/codex/guides/agents-md
What is the American engineer and German engineer comparison?: Peter Steinberger, who created OpenClaw, told Lex Fridman that Opus was a little bit too American, and Lex answered that Codex is German. I have heard it applied to Claude Code and Codex: https://lexfridman.com/peter-steinberger-transcript
How should you compare Codex and Claude Code?: Use a greenfield project. Add skills and an AGENTS.md that match the project, and run both tools on it. A project where you already use one tool deeply makes the comparison hard.
What changes for Codex usage in December?: OpenAI says Plus, Pro and Business plans share one usage allowance across ChatGPT and Codex from December, measured in tokens: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits