Codex Pricing: Plans, Credits and API-Key Billing

OpenAI's Codex has no separate price. It comes with ChatGPT: Free, Go at $8, Plus at $20, Pro from $100, and Business at $20 a seat billed annually. Developers can instead run it with an API key and pay per token. Usage is shared with ChatGPT Work, and extra use is bought as credits. Prices checked 2026-10-31.

Prices checked 2026-10-31

Source: OpenAI's Codex pricing page, plus the API pricing page for token rates. The ChatGPT pricing page itself blocks automated fetches, so we used OpenAI's developer docs, which show the plan prices embedded in the Codex page.

For the agent itself, see OpenAI Codex CLI. For OpenAI's token prices, see OpenAI API pricing. For other tools, see AI coding agent pricing compared.

What each route gets you

ChatGPT plans (Free, Go, Plus, Pro)

For Plus, OpenAI lists Codex on the web, in the CLI, in the IDE extension and on iOS, plus cloud integrations such as automatic code review and Slack, access to GPT-6.1 Sol and GPT-6 Luna, and the option to extend usage with ChatGPT credits. Pro adds larger plans at $100, $200 or $500 a month, and Ultrafast mode on the $500 plan.

The Free and Go listings are shorter. They mention GPT-6 Luna at Standard speed in the desktop app, subject to rollout. Check the plan card for the detail on those tiers.

Business, Enterprise and Edu

Business adds access across desktop and mobile apps, larger virtual machines for cloud chats, a dedicated workspace with essential admin controls, SAML SSO and MFA, and no training on your business data by default. Enterprise adds priority request processing, SCIM, EKM, user analytics, domain verification, role-based access control, audit logs and usage monitoring through a Compliance API, and data retention and residency controls.

API key

For automation in shared environments like CI, you can sign in with an API key. You get Codex in the CLI, SDK or IDE extension. There are no cloud-based features, so no GitHub code review and no Slack integration. Model availability follows the API models your key can use, and you pay for Codex usage at API pricing. OpenAI notes that API token prices are separate from subscription usage and should not be used to estimate how many tasks a plan includes.

How usage limits work

OpenAI says "ChatGPT Work and Codex share usage": Work in ChatGPT draws on the same pricing, credits and limits as Codex. A message allowance depends on the model, the size and complexity of the task, and whether you run it locally or in the cloud. Tasks that look similar can consume different amounts. OpenAI says prompt length alone is not a reliable estimate.

For Plus and Standard Business, OpenAI gives estimates of local messages per five-hour period. These are not fixed limits.

Those ranges are wide for a reason. They depend on the task. They are also the only published numbers, so treat them as order-of-magnitude. Pro plans currently have no five-hour limit. Cloud tasks may use more of your allowance than local messages, and weekly limits may also apply. Your usage dashboard shows current limits and reset times, and /status shows remaining limits in a Codex CLI session.

Other items share the same allowance. Image generation uses included limits about 3 to 5 times faster on average. ChatGPT for Excel on Plus and Pro shares usage once pricing for it takes effect. GPT-6.1 Sol usage drawn from a subscription has no long-context surcharge in Codex or ChatGPT Work.

Speed modes

Fast and Ultrafast modes burn more of your allowance.

Credits

When you reach your limits, credits let you continue. Plus and Pro users can buy additional credits without upgrading. Business, Edu and Enterprise plans with flexible pricing can buy workspace credits. If you hit a limit mid-turn, the agent can usually finish that turn, subject to fair use limits. You may also run extra local chats with an API key, at standard API rates.

Codex credit billing has no separate cache-write charge. Credit rates per million tokens at Standard speed:

OpenAI's page does not state the dollar price of a credit on this page; purchase prices and discounts depend on your plan or agreement. Comparing the credit ratios with the API prices in dollars, GPT-6.1 Sol is 50 credits against $2 per million input tokens, which is consistent with 25 credits to the dollar, but we have not confirmed that. The page says a typical GPT-5.6 Sol task may use 2 to 15 credits, and that GPT-5.6 Sol's promotional pricing is available at least through 21 November 2026.

Other costs on the page

• Voice in the desktop app uses your Codex budget at $0.05 per minute. Credit-based workspaces pay 1.25 credits per minute. The model doing the task is billed separately.
• Code Review usage applies only when Codex runs reviews through GitHub, for example when you tag @Codex on a pull request or enable automatic reviews. Reviews run locally count toward general limits.
• Sites is included with eligible plans during public beta.
• Referrals. Eligible users can invite friends or coworkers from the profile menu. Rewards and eligibility vary by plan and promotion, and are not available for Enterprise.

A model retirement to know about

GPT-5.5 retires from Codex with ChatGPT sign-in on 14 October 2026 at noon PDT (19:00 UTC). The OpenAI API and Codex sessions authenticated with an API key are not affected. If you still reference GPT-5.5 in scripts, check the model guidance.

Privacy and admin by plan

The pricing page is explicit about two things. Business plans have no training on business data by default, and Enterprise adds data retention and data residency controls, EKM and a Compliance API for audit logs. For personal plans the page does not state a training default, so check OpenAI's data controls before you use Codex on private code.

GPT-6.1 Terra does not support UAE inference residency. GPT-6.1 Sol is off by default for Enterprise and Edu until an administrator enables it, and it is available in Work and Codex, not in Chat.

Moving to Codex from another agent

If you are switching, three differences change your costs.

1. You are buying ChatGPT too. The plan includes ChatGPT features, so part of the price covers things you may not use. If you only want a coding agent, the API-key route pays for tokens alone.
2. Usage is shared. Work and Codex share one allowance, as Claude's chat and Claude Code do. See Claude Code pricing for that model.
3. Context files matter. OpenAI's tips for making limits last mention shrinking AGENTS.md and limiting MCP servers, because each adds context to every message. Codex reads AGENTS.md, so a long one has a recurring cost.

Cursor takes yet another route, with a weekly pool and prepaid credit; see Cursor pricing.

How to estimate before you buy

Run Codex on the free tier or with a small API budget for a week on real tasks. Note which model you used, then look at the usage dashboard. If you stayed well inside the Plus window, buy Plus. If you hit the wall twice in a week, price Pro. If your work is scripted, count tokens on the API key route and compare with the plan. Do not extrapolate from API token prices to subscription allowances. OpenAI says plainly not to.

ChatGPT or API key?

A rule of thumb from the arithmetic: if you can predict your tokens, the API is easy to budget; if you cannot, a plan caps your exposure. Using API prices, one request with 50,000 input and 2,000 output tokens on GPT-6.1 Sol is $0.12, on GPT-6.1 Terra $0.06 and on GPT-6 Luna $0.006. A Plus plan at $20 covers about 167 of those Sol requests at API prices. That is a comparison of prices, not of Plus allowance, which OpenAI describes only as message ranges.

Who should pick which

Our take

OpenAI built Codex pricing as part of ChatGPT, and that shows. The tiers are clear but the allowance is not: message ranges as wide as 5 to 45 tell you the real limit depends on your work. The upside is that Pro has no five-hour wall. The downside is that you cannot forecast precisely.

If you want predictable spend, use the API key route and meter it. If you want a flat price, pick the smallest plan that survives a week and set up credits as the overflow. We have not tested Codex, so we say nothing about quality, only how it is priced. Compare it with Claude Code pricing, Cursor pricing and GitHub Copilot pricing.

Checklist

• Decide whether you need cloud features such as GitHub review.
• Check your usage dashboard every week or two.
• Trim AGENTS.md and disable unused MCP servers; both add context to every message.
• Use Terra or Luna for small tasks. Their message ranges are far higher.
• For CI, use an API key and set API spend limits.
• Confirm plan prices on chatgpt.com/pricing, which we could not fetch.

Practical limits to plan around

Because OpenAI describes allowances as message ranges, plan around the low end of each range, not the high end. On Plus, GPT-6.1 Sol is listed at 15 to 160 local messages per five hours, so a long agentic task may behave like the low figure. Switching routine edits to GPT-6.1 Terra (35 to 285) or GPT-6 Luna (350 to 3,000) stretches the allowance by an order of magnitude, if the smaller model can do the job.

Cloud tasks cost more than local messages, and Fast mode multiplies included usage by 2.5, so avoid Fast for background work. Image generation also drains the allowance about three to five times faster than a similar turn without it. If you use Codex for code review through GitHub, remember that tagging @Codex on pull requests or enabling automatic reviews is the usage the page counts as Code Review.