Codex costs what your ChatGPT plan costs; there is no separate Codex price. According to OpenAI's pricing page, as of October 26, 2026, Free (0 dollars) and Go (8 dollars a month) only get the desktop app with the small model GPT-6 Luna, subject to rollout. From Plus (20 dollars) up you get the command line, the editor extension, the web and iOS. Pro comes in three tiers at 100, 200 or 500 dollars a month, currently without a five-hour limit but with possible weekly limits. Business costs 20 dollars per user a month on annual billing with at least two users, 25 dollars billed monthly. When your allowance runs out, you buy credits on Plus and Pro or work with an API key at API prices. For comparison: Claude Code comes with Claude Pro at 20 dollars a month, and Max costs 100 or 200 dollars.
I work with Claude Code every day, my AI workforce runs on it, and I use Codex regularly for correction loops: one tool builds, the other checks. I have not measured Codex usage myself. All figures here come from OpenAI's pricing page and models overview, retrieved on October 26, 2026, and for Claude from Anthropic's pricing page on the same day. What sets the three coding agents apart beyond price is in Claude Code, Codex or Copilot: Which One?. Everything I write about Claude is collected on the Claude page.
Codex pricing 2026: the plans at a glance
OpenAI lists the plans on its Codex pricing page. Prices in US dollars per month, as of October 26, 2026:
Since October 26, 2026 there is also GPT-6.1 Terra, which per the changelog is available from Plus up and only in Codex and ChatGPT Work. One thing matters for choosing a plan: ChatGPT Work, the Codex technology inside chat, draws on the same allowance. OpenAI puts it as: "ChatGPT Work and Codex share usage." If you use Work a lot, you have less left for Codex.
How OpenAI counts usage limits
OpenAI does not give fixed message counts, but estimates of local messages per five-hour window, for Plus and the standard Business seat. Per the pricing page:
The ranges are wide because usage depends on the model, the size of the task, context and tools, per OpenAI. In its words: "These estimates are not fixed message limits". Cloud tasks use more than local messages, according to OpenAI, and the note under the table says "Weekly limits may also apply". For Pro: "Pro plans currently have no five-hour limit." Both the "currently" and the weekly-limit note are there, so Pro is not "unlimited".
A few things eat into the allowance faster, per the pricing page:
- Fast mode: 2.5 times the usage from your subscription allowance.
- Ultrafast: 8 times the usage, only on Pro 500 and eligible Enterprise and Edu plans.
- Image generation: on average three to five times faster than similar turns without an image.
- Voice in the desktop app: 0.05 dollars per minute from your Codex budget.
You can see where you stand in the usage dashboard in your Codex settings, and in the command line with /status. OpenAI itself recommends: "Check the dashboard every week or two to understand your pace and remaining capacity." And if you hit the limit mid-task, the agent can finish the current turn, per OpenAI, subject to fair use limits.
Credits: what happens when your allowance runs out
Credits are the unit OpenAI uses to bill usage beyond your plan. Per the pricing page, Plus and Pro users can buy credits instead of upgrading; Business, Enterprise and Edu with flexible pricing buy credits for the whole workspace. What a credit costs in dollars is not on the pricing page, which only says "Credit purchase prices and applicable discounts depend on your plan or agreement". The help articles OpenAI links to returned an error to me on October 26, 2026 (HTTP 403). So I leave the dollar price per credit open.
What the pricing page does give is the usage in credits per million tokens, at standard speed:
This shows how much the model choice matters: according to this table, Astra costs five times as much per input and output token as GPT-6.1 Sol, Luna a twentieth. If you buy credits, the model is where you save. The third route is an API key: then you pay OpenAI's API prices, independent of your ChatGPT plan, but without cloud features.
Codex vs Claude Code: costs compared
According to both vendors' pricing pages, as of October 26, 2026:
At entry level, both cost the same. Above that the ladders diverge: OpenAI has three Pro tiers, whose differences in limits I could not verify, and currently no five-hour limit on Pro, Anthropic has two Max tiers that, per its pricing page, give five or twenty times the usage per window of Pro. How much work you get for your money cannot be read off the pricing pages, because both meter by task, model and context. I have not run my own usage comparison; any claim that "X lasts longer than Y" would be a guess.
From my experience with Claude Code I can only describe the path: I did not start on Max, I moved only when Pro's allowance stopped covering long sessions. The same path works for Codex: start small, check the dashboard after a week or two, as OpenAI itself recommends, and then decide. Which other alternatives to Claude Code exist and what they cost is covered in Claude Code Alternatives in 2026.
What works in the EU
I use Codex itself in Germany. Some features are restricted, though, per OpenAI as of October 26, 2026:
- Dots, the always-on agents on GPT-6 Astra: on Pro 100, 200 and 500 only "for users over 18 outside the European Economic Area, United Kingdom, and Switzerland", per the dots documentation. Business Premium and Enterprise are rolling out worldwide. Per OpenAI, dots do not support data or inference residency. Anthropic's counterpart is covered in Claude Duet: What Anthropic's Agent Does.
- Computer Use and Record & Replay: per the plan comparison "currently limited to only specific regions"; the pricing page does not say which.
- Composer predictions (suggestions for your next message): beta, personal Pro plans only, "in supported regions" per the changelog; OpenAI does not say whether the EU is one of them.
- Data residency and retention controls appear in the plan comparison for Enterprise and Edu only. For Ultrafast with GPT-6.1 Sol, OpenAI names inference residency in the United States or Europe (EEA and Switzerland).
- Training: "No training on API or business data by default" is listed for Business, Enterprise and API keys, not for Plus and Pro.
This isn't legal advice. For company code and client data, use a business plan, not a personal subscription; whether that is enough for your case is something to settle with a qualified professional. How this looks on the Claude side is in Claude Privacy and GDPR: What Applies.
Which plan for whom
My assessment based on the pricing pages, not on a test:
- You just want to look at Codex: Free or Go are enough for a first impression in the desktop app. For the command line you need Plus.
- You use Codex as a checker next to Claude Code: start with Plus and read the dashboard after a week or two. A higher tier only pays off if you hit the limit regularly.
- You work with Codex all day: one of the Pro tiers. Per the pricing page, Pro 500 is the only personal tier with Ultrafast. How Pro 100, 200 and 500 differ in limits is in a help article that returned an error to me on October 26, 2026 (HTTP 403), so I leave that open.
- Your team works with client data: Business, for SSO and no training by default; Enterprise if you need data residency.
- You want to run Codex in scripts or CI: an API key, billed at API prices.
If you use Codex and Claude Code side by side, keep your house rules lean, too. On its pricing page OpenAI explicitly advises keeping your AGENTS.md small, because it brings context into every session. How AGENTS.md and CLAUDE.md relate is in CLAUDE.md vs AGENTS.md: What Survives.
Frequently asked questions
Is Codex free?
Partly. Per OpenAI, Codex is included in Free and Go, but only in the ChatGPT desktop app with GPT-6 Luna, subject to rollout. OpenAI lists the command line, the editor extension and the web from Plus up, at 20 dollars a month.
How much does Codex Pro cost?
Per OpenAI, Pro comes in three tiers at 100, 200 and 500 dollars a month. All three currently have no five-hour limit per the pricing page; weekly limits are possible. Only Pro 500 includes Ultrafast mode. How the three tiers differ in limits is in an OpenAI help article I could not retrieve (HTTP 403).
Does Codex Pro have a usage limit?
OpenAI currently names no five-hour limit for Pro, in its words "Pro plans currently have no five-hour limit". The pricing page does not rule out weekly limits, though. You can see your usage in the usage dashboard or with /status in the command line.
How much does a Codex credit cost?
The Codex pricing page does not give a dollar price per credit; per OpenAI it depends on your plan and agreement. What is documented is the usage per million tokens, for example 50 credits input and 250 credits output for GPT-6.1 Sol.
Is Codex cheaper than Claude Code?
Not at entry level: Plus and Claude Pro both cost 20 dollars a month. Above that, OpenAI has more tiers (100, 200, 500 dollars) than Anthropic (100, 200 dollars). How much work you get per dollar depends on task and model for both; there is no reliable answer without your own measurement.
Can I use Codex without a ChatGPT plan?
Yes, with an API key. Codex then runs in the command line, the SDK and the editor extension, billed at OpenAI's API prices, but without cloud features such as GitHub code review or the Slack integration.
Where to go next
If you are still deciding between the tools, start with Claude Code, Codex or Copilot: Which One?. If your team works in GitHub, add Claude vs Copilot for Coding: Which One?. And if you decide on Claude Code, begin with Claude Code: A Guide for Beginners. How I run a whole workforce on Claude Code is what I show in my community.
