Last updated: 6 October 2026
Quick Answer: Codex pricing is a ChatGPT subscription, not a separate product price: every paid plan includes Codex, so the plan you pick is the bill. Plus is the entry point for real work, Pro removes the five-hour usage window, Enterprise is quoted by sales, and an API key is metered per token instead.
The plan price is the floor. Usage limits and credits decide what a team pays.
OpenAI Codex pricing is not a product price. Codex is bundled into ChatGPT subscriptions, so you buy a plan and get an allowance measured in messages and credits. Two teams on identical plans can run very different bills, because model, context length and cloud execution all draw on it.
How much does OpenAI Codex cost in 2026?
Codex has no price of its own. It is included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise, so the plan you pick is the bill. What separates them is where Codex runs and how much it does.
On the individual ladder, Free is $0, Go is $8 a month, Plus is $20, and Pro starts at $100 with tiers at $100, $200 and $500, the top adding Astra Ultrafast, per OpenAI's Codex pricing page. Enterprise and Edu carry no list price.
Codex pricing plans as listed on learn.chatgpt.com on 6 October 2026; prices and links in the text.
| Plan | Where Codex runs | Usage |
|---|---|---|
| Free and Go | Desktop app only, subject to rollout | Limited access, GPT-6 Luna |
| Plus | Web, CLI, IDE extension, iOS | Estimates per five-hour window |
| Pro | Everything in Plus, adds Astra Ultrafast | No five-hour limit |
| Business | Desktop, mobile, web; larger cloud machines | Same estimates as Plus |
| Enterprise and Edu | Everything in Business | No fixed limits on flexible pricing |
Is Codex free with ChatGPT, and which plans include it?
Codex is not free in any sense a team can use. Free and Go accounts get GPT-6 Luna in the desktop app only, and OpenAI marks that as subject to rollout. Every surface a developer needs, the CLI, the IDE extension and the web, starts at Plus.
That is where ChatGPT Codex pricing misleads people: the plan page pitches Free at "quick coding tasks", Plus at "focused coding sessions", which reads like a volume difference when it is also a surface one. ChatGPT Work shares the same allowance, so budget for that overlap.
What are the Codex usage limits on Plus, Pro and Business?
The limits are estimates, not caps. OpenAI publishes local messages per five-hour period, identical on Plus and Standard Business: GPT-6 Astra 5 to 45, GPT-6.1 Sol 15 to 160, GPT-6 Sol 15 to 150, GPT-6 Luna 350 to 3,000. Weekly limits may also apply.
The ranges are wide because a message is not a unit of work: model, context length, reasoning and tool calls all change what one consumes. Fast mode spends included usage at 2.5 times the Standard rate.
How much does Codex Pro cost, and is it unlimited?
Codex Pro pricing runs across three monthly tiers, and only the top carries Astra Ultrafast. None is unlimited: OpenAI's wording is Pro plans currently have no five-hour limit, which removes the window and leaves fair-use limits in place. Hit a limit on Plus or Pro and you buy extra credits rather than upgrade, though what a credit costs is unpublished.
How is Codex billed through the API and the Codex CLI?
With an API key, Codex is billed per token and loses its cloud features. Codex API pricing is published per model: one model is Codex-specific, and anything else the CLI calls bills at its ordinary rate.
What does Codex cost per token on the API?
Codex has one dedicated API model. Per million tokens at Standard speed, gpt-5.3-codex costs $1.75 input, $0.175 cached and $14.00 output; Fast doubles that to $3.50, $0.35 and $28.00. For comparison GPT-6.1 Sol runs $2.00 in and $10.00 out, and GPT-6 Luna $0.10 and $0.50, per OpenAI's API pricing.
Two things to watch. The older codex-mini and gpt-5-codex models are no longer listed, so do not budget against them. Long context above 272K input tokens roughly doubles the rate, and residency endpoints add a 10 percent uplift.
Is the Codex CLI free?
The CLI itself is free, open source under Apache-2.0 in OpenAI's codex repository. Codex CLI pricing is really a question about sign-in: sign in with ChatGPT and your plan's allowance applies, or use an API key and pay per token. The API-key route has no cloud features, so GitHub code review and Slack stop working.
What do Codex Business and Enterprise plans add for teams?
Business adds the workspace, not the capability: a dedicated workspace with essential admin controls, SAML SSO, MFA and cloud-managed config policies, and no training on your business data by default. It starts at two users. Codex enterprise pricing is unpublished; OpenAI quotes it.
Enterprise carries the controls a security review asks about: SCIM, Enterprise Key Management, role-based access control, domain verification, user analytics, audit logs through the Compliance API, and data retention and residency controls. Read that one closely. Residency governs where OpenAI processes your data, not whether your code leaves your network, and Codex stays a hosted service. If source code must never cross your own boundary, residency does not satisfy it.
What would a 50-developer team pay for Codex each month?
Fifty developers on Business seats come to $1,000 a month billed annually, or $1,250 monthly, at the $20 and $25 per-user rates on OpenAI's Business and Enterprise plans.
Treat that as the floor, not the forecast. Three things sit on top:
- Credits. Usage past the allowance draws on workspace credits. The credits-per-million-tokens rates are published; the dollar price per credit is not.
- Heavy users. A few developers in long sessions take a disproportionate share, and moving them to Pro is a separate cost.
- Shared usage. ChatGPT Work draws on the same pool, leaving less for coding.
How do you keep Codex spend predictable?
Predictability comes from controls, not forecasts. Codex has no budget cap to set, so read the usage dashboard every week or two, as OpenAI advises. Route the API-key half through a self-hosted LLM gateway so every request carries a team and a project, and budgets bite before spend happens.
Is Codex cheaper than Claude Code?
At the entry point the two are level, so price is not the deciding factor. Claude Pro is $17 a month billed annually or $20 monthly, Max starts at $100, a standard Team seat $20 annually or $25 monthly, a premium Team seat $100 or $125, and Enterprise $20 for each seat annually plus usage at API rates, per Anthropic's pricing page.
The real difference is metering. Codex gives one estimated message allowance per five-hour window that all your ChatGPT usage shares, topped up with credits that expire. Claude Code attaches an allowance to each seat and, on Enterprise, meters every token at API rates. A team with a few heavy users fares quite differently under those designs, which is why codex vs claude code pricing rarely resolves on the plan page. See our Claude Code pricing breakdown and the Codex and Claude Code comparison.
What mistakes should you avoid when budgeting for Codex?
Nearly every error comes from treating an allowance like a seat.
- Budgeting from the plan price alone. The plan is the floor; credits and heavy users are the variable.
- Reading the message estimates as caps. They are ranges that move with model, context and tool use.
- Forgetting the shared pool, so ChatGPT Work usage quietly eats the coding allowance.
- Planning around a model that is leaving. GPT-5.5 retires from Codex on all plans on 14 October 2026.
How Origins AI approaches this
If you are modelling this spend because nobody can say who uses which model, on which project, at what cost, that is a gateway problem, not a plan problem. Origins AI (originshq.com) builds the Origins AI Coding Tool, described on its product page as a self-hosted AI coding assistant and LLM gateway for your own infrastructure.
According to that page, read on 6 October 2026, the gateway proxies LLM requests from engineering tools behind an OpenAI-compatible REST API, with routing, team quotas, role-based access control and cost tracking, and the company reports that it tracks every LLM request by team, project and engineer. It is deployed rather than subscribed to: the page lists on-premise, private cloud in your own AWS, Azure or GCP account, hybrid, and air-gapped modes running local models. In on-premise and air-gapped modes the page states that no source code is sent to any external service; in hybrid mode, only the code context submitted to the model leaves your network. The company does not publish a rate card. The Coding Tool product page has the architecture.
Talk to an engineer
Send your team size and the AI coding tools your developers use, and we will map the cost and the controls. Book a call.