Codex Pricing: Plans, Usage Limits and API Cost Math

Codex pricing starts at $0. As of October 2026, OpenAI's Codex coding agent is part of every ChatGPT plan. That means Free ($0), Go ($8 a month), Plus ($20), Pro ($100, $200 or $500), Business ($20 to $25 per user) and Enterprise.

The catch is that Free and Go only get the small GPT-6 Luna model in the desktop app. The full Codex (CLI, IDE extension, web, cloud tasks and GPT-6.1 Sol) starts at Plus.

You can also skip ChatGPT and run Codex with an OpenAI API key, paying per token. That's cheap on small models and expensive on the top one, as the math below shows.

How we researched this: OpenAI's Codex pricing and models docs, the OpenAI API pricing page and the Codex GitHub repo as of October 2026, plus press coverage where OpenAI doesn't publish a number. We haven't run our own benchmarks, so the cost examples are calculations from list prices, not measured bills.

Codex pricing by plan

Here is every way to pay for Codex, from OpenAI's Codex pricing page, checked October 25, 2026.

A few details matter more than the sticker price:

• OpenAI's own one-liner for Plus is "a few focused coding sessions each week." That's an honest hint that Plus is not a full-time plan.
• Local and cloud work share one allowance. OpenAI says local messages and cloud chats draw from the same pool, and cloud tasks may use more of it.
• Model matters a lot. A Plus user gets roughly 20 times more GPT-6 Luna messages than GPT-6.1 Sol messages (see the limits below).
• GPT-5.5 is gone from plan sign-in. It retired from Codex with ChatGPT sign-in on October 14, 2026. API-key sessions weren't affected.

What are the Codex usage limits?

OpenAI publishes ranges, not fixed numbers. For Plus and Standard Business seats, these are the local messages per five-hour window from the pricing docs, as of October 2026:

The range is wide because a "message" is one prompt you send, and one prompt can trigger a few model calls or a hundred. A big repo, a long session and a high effort setting all push you toward the low end. OpenAI adds that "weekly limits may also apply" but doesn't publish them.

Speed modes burn the allowance faster. On included plan usage, Fast mode costs 2.5x and Ultrafast (Astra or GPT-6.1 Sol) costs 8x the Standard rate. Image generation uses limits about 3 to 5 times faster.

Pro plan limits

OpenAI says Pro plans "currently have no five-hour limit," but its pricing page doesn't say how much more usage each Pro tier gets.

Press coverage fills part of the gap. The $100 tier launched in April 2026 with 5x Plus usage. One blog report says the $200 tier drops from 20x to 10x Plus on October 29, 2026 for existing subscribers. OpenAI's pricing page doesn't confirm the multiples, so treat that as unconfirmed.

Reports put the $500 tier at 25x Plus, but at least one outlet notes OpenAI never stated that number. Treat all three multiples as reported, not official.

If those reports hold, the $200 tier is now the weakest value per dollar: twice the price of the $100 tier for twice the usage, while $500 buys 2.5x the $200 tier plus Ultrafast.

What happens when you hit the limit?

You don't get cut off mid-task. OpenAI says the agent can finish the turn it's on when you hit a limit, subject to fair use. After that you have four options:

1. Wait for the five-hour window to reset.
2. Buy ChatGPT credits (Plus and Pro) without upgrading your plan. Business, Edu and Enterprise workspaces on flexible pricing buy workspace credits.
3. Switch to an API key and keep working locally at API rates. Any user can do this.
4. Upgrade to Pro, which removes the five-hour window.

To check where you stand, run /status in a Codex CLI session or open the usage dashboard.

How do Codex credits work?

Credits are how OpenAI bills extra usage beyond your plan. They're priced per million tokens, like the API. Here are the Standard-speed rates from the pricing docs, as of October 2026:

OpenAI says a typical GPT-5.6 Sol task uses 2 to 15 credits. We couldn't load the help article with the dollar price of a credit. One clue: every credit rate above is exactly 25 times the matching API dollar price, which would make one credit worth about $0.04. That's our inference, not a published price, so check the price on your own billing page before you buy.

What does Codex cost with an API key?

With an API key, Codex bills at standard API prices instead of plan limits. These are the per-million-token prices for short prompts (up to 272K input tokens) from the OpenAI API pricing page, as of October 2026:

Two things to watch. Above 272K input tokens, long-context rates kick in (GPT-6.1 Sol goes to $4 input and $15 output). And Fast mode on the API costs double the Standard rate, while Ultrafast GPT-6.1 Sol costs six times as much ($12 input, $60 output).

The API key route has real limits beyond price. You lose cloud tasks, GitHub code review and Slack, and the Codex GitHub README says API-key use "requires additional setup." OpenAI's own recommendation there is to sign in with a ChatGPT plan.

Worked examples: API key vs subscription

To compare the options we need a unit. We use one "task": a medium bug fix or small feature that takes the agent about 30 model requests (reading files, editing, running tests, fixing). These are our assumptions, the same ones we used in our Claude Code pricing guide, not OpenAI numbers:

• Each request carries about 55,000 tokens of cached context (read from cache).
• Each request adds about 5,000 new tokens that get written to the cache.
• Each request produces about 1,500 output tokens, including reasoning.

Here is the math for one task at October 2026 API list prices.

GPT-6.1 Sol, one task:

• Cache reads: 30 × 55,000 = 1.65M tokens × $0.10 = $0.165.
• Cache writes: 30 × 5,000 = 150,000 tokens × $2.50 = $0.375.
• Output: 30 × 1,500 = 45,000 tokens × $10 = $0.45.
• Total: about $0.99.

GPT-6 Astra, one task: $1.65 + $1.875 + $2.25 = about $5.78. Astra costs roughly six times as much as Sol for the same work.

GPT-6 Luna, one task: about $0.06. That's why Plus gets thousands of Luna messages but only dozens of Astra ones.

On credits, the same GPT-6.1 Sol task works out to about 23 credits. The credit table has no cache-write rate, so we counted the new tokens as input; treat that figure as a rough guide.

Now scale it to a month:

The takeaway: if your work fits inside a plan, the subscription is far cheaper than an API key on GPT-6.1 Sol, and the gap is huge on Astra. The "if" is the problem. OpenAI doesn't publish limits in tokens, so we can't say whether 200 tasks fit in Pro $100. The practical test is to start on a plan and watch /status for a week.

The API wins in two cases. Scripts and CI jobs, where per-token billing is simpler than a seat. And Luna-sized jobs: 660 small tasks on GPT-6 Luna come to about $38 a month.

Which ChatGPT plan should you pick for Codex?

Pick by how often you'd hit limits, not by features. From Plus up, every plan gets the same Codex tool.

• Free or Go fits trying Codex on small tasks. You only get GPT-6 Luna in the desktop app, so no CLI and no GPT-6.1 Sol. Go adds nothing for Codex over Free, so don't buy Go for coding.
• Plus ($20) fits a few sessions a week, in OpenAI's own words. At API rates, about 20 of our GPT-6.1 Sol tasks would already cost $20, so Plus pays off quickly if the limits hold for you.
• Pro $100 fits anyone using Codex most workdays. It drops the five-hour window, which is usually the first wall Plus users hit.
• Pro $200 is hard to recommend after the reported cut to 10x Plus. Pick it only if the $100 tier stops you and $500 is out of budget.
• Pro $500 fits heavy agent users and anyone who needs Ultrafast mode, which only this tier gets. Ultrafast drains usage 8x faster, so it's for short bursts.
• Business fits teams of two or more who want admin controls, SSO and no training on their code by default. At $20 to $25 a seat, it costs about the same as Plus.
• An API key fits CI, the Codex SDK and overflow work when your plan runs out. Don't use it as your main way to run Astra unless someone else is paying.

Is Codex free?

Yes, in a limited form. ChatGPT Free includes Codex with GPT-6 Luna in the desktop app, "subject to rollout." OpenAI doesn't publish the Free usage limits. The CLI, IDE extension, cloud tasks and GPT-6.1 Sol all need Plus or higher, or an API key.

The Codex CLI itself is open source under the Apache-2.0 license, so the software is free; you pay for the model. For other no-cost setups, see our guide to the best free AI for coding.

Codex vs Claude Code cost

At the entry level they cost the same. Codex Plus and Claude Pro are both $20 a month, and our standard task costs about $0.99 on both GPT-6.1 Sol and Claude Sonnet 5.5 at API rates, since their list prices match.

The differences are at the edges. Codex has a real free tier and an $8 Go plan; Claude Code has neither. Codex Pro removes the five-hour window, while Claude's Max plans keep one.

On the other hand, Claude Max 20x at $200 is a clearer deal than Codex Pro $200 after its reported cut. For the full feature comparison, see Claude Code vs Codex, and for Codex against an AI editor, Codex vs Cursor.

How to make your Codex plan last longer

Most of these come from OpenAI's pricing docs:

1. Use GPT-6.1 Sol by default and Luna for small, repeatable jobs. Save Astra for the hard ones; each Astra message costs several Sol messages.
2. Leave Fast and Ultrafast off unless you need speed. They cost 2.5x and 8x your included usage.
3. Keep AGENTS.md small. It loads with every task. OpenAI suggests nesting smaller files inside the repo for big projects.
4. Turn off MCP servers you aren't using. Each one adds context and eats usage.
5. Point Codex at the files that matter and say what "done" looks like. Vague prompts mean more model calls.
6. Check /status or the usage dashboard every week or two, as OpenAI suggests.

FAQ

How much does Codex cost?

As of October 2026, Codex is included in ChatGPT Free ($0), Go ($8), Plus ($20), Pro ($100, $200 or $500), Business ($20 to $25 per user) and Enterprise. With an API key, you pay per token, for example $2 input and $10 output per million tokens on GPT-6.1 Sol.

How much does Codex CLI cost?

The CLI is free and open source. To use it you need ChatGPT Plus or higher, or an OpenAI API key billed per token. Free and Go plans only get Codex in the desktop app.

What are the Codex usage limits on Plus?

On GPT-6.1 Sol, Plus gets 15 to 160 local messages per five hours; on GPT-6 Luna, 350 to 3,000. OpenAI says weekly limits may also apply but doesn't publish them.

What happens when I run out of Codex usage?

The current turn can finish. Then you wait for the reset, buy credits (Plus and Pro), switch to an API key, or upgrade. Pro plans currently have no five-hour limit.