OpenAI Codex Pricing (October 2026): Which ChatGPT Plans Include Codex

Codex, OpenAI's coding agent, doesn't have a separate subscription. It comes with every ChatGPT plan and shares one usage allowance with ChatGPT Work. As of October 17, 2026, the plans are Free ($0), Go ($8), Plus ($20), Pro ($100, $200 or $500), Business ($25 per user, or $20 annual) and Enterprise/Edu (custom). The other route is an API key, which you pay for per token at OpenAI API rates. This page covers what each option includes, how the five-hour limits and credits work, and what typical coding sessions cost.

What each plan gets in Codex

Free and Go are aimed at small, quick jobs: explaining a file, writing a function, fixing a small bug. OpenAI's models page says GPT-6.1 Sol rolled out to Plus, Pro, Business, Enterprise and Edu, not to Free or Go. For real day-to-day use in a repository, Plus is the starting point.

How the five-hour limits work

On Plus and standard Business seats, usage is counted in rolling five-hour windows, with weekly limits on top. OpenAI publishes estimated ranges of local messages per five hours rather than hard numbers:

The wide ranges matter. A message that asks Codex to refactor a large module reads many files, runs tests and writes lots of code, using far more of your allowance than a one-line question. Larger repositories, longer sessions and higher reasoning effort all push you toward the low end. You can see current usage and reset times in the usage dashboard, or run /status in Codex CLI.

Pro 100, Pro 200 and Pro 500 currently have no five-hour limit, but weekly limits may apply. The three Pro tiers differ in how much is included. Pro 500 has the most and is the only one with Ultrafast. Pro 200 reopened on September 29, 2026 with a smaller allowance, and subscribers who already had it keep the old allowance through October 29, 2026 (see ChatGPT pricing for details).

Credits: paying for extra Codex usage

When you run out of included usage, Plus and Pro users can buy ChatGPT credits, and Business, Enterprise and Edu workspaces on flexible pricing can buy workspace credits. Credits are charged per million tokens, by model:

Each rate is exactly 25 times the matching API dollar price. For example, GPT-6.1 Sol is $2 per 1M input tokens on the API and 50 credits in Codex. OpenAI's pricing page doesn't state a dollar price per credit, but OpenAI's student offer describes $100 as 2,500 credits, or $0.04 per credit. At that value, credit usage costs the same as the API. Your account shows the actual purchase price, and workspace prices depend on your plan or agreement.

• Fast mode costs 2x the Standard credit rate when paid from credits, and 2.5x when it uses included subscription usage.
• Ultrafast (GPT-6 Astra and GPT-6.1 Sol only) costs 6x from credits and 8x from included usage.
• Desktop voice costs $0.05 per minute from your Codex budget on personal plans, and 1.25 credits per minute on credit-based Business, Edu and Enterprise workspaces.
• If you hit your limit during an active turn, Codex can finish that turn, subject to fair use limits.

Worked cost examples

Coding agents resend a lot of context, so most input is usually cached. These examples assume one working session sends 2M input tokens, 1.6M of them cached, and produces 60K output tokens. Dollar figures for credits use the $0.04 per credit value from OpenAI's student offer.

1. A session on GPT-6.1 Sol

Uncached input: 0.4M × 50 = 20 credits. Cached input: 1.6M × 2.5 = 4 credits. Output: 0.06M × 250 = 15 credits. Total: 39 credits, about 39 × $0.04 = $1.56. With an API key the same session costs 0.4 × $2 = $0.80, plus 1.6 × $0.10 = $0.16, plus 0.06 × $10 = $0.60, also $1.56.

2. The same session on GPT-6 Astra

Uncached input: 0.4M × 250 = 100 credits. Cached input: 1.6M × 25 = 40 credits. Output: 0.06M × 1,250 = 75 credits. Total: 215 credits (about $8.60), which is 5.5 times the Sol session. In Fast mode paid from credits it would be 215 × 2 = 430 credits, and in Ultrafast 215 × 6 = 1,290 credits (about $51.60).

3. The same session on GPT-6 Luna

Uncached input: 0.4M × 2.5 = 1 credit. Cached input: 1.6M × 0.25 = 0.4 credits. Output: 0.06M × 12.5 = 0.75 credits. Total: 2.15 credits, under $0.09. That is why Free and Go users get Luna and why Plus users get up to 3,000 Luna messages per five hours.

4. Plus plus credits, or Pro 100?

Pro 100 costs $100 - $20 = $80 a month more than Plus. At about $0.04 per credit, $80 buys 2,000 credits, or roughly 2,000 ÷ 39 ≈ 51 extra GPT-6.1 Sol sessions of the size above. If you regularly need more than that on top of your Plus allowance, Pro 100 is likely cheaper and also removes the five-hour limit. If you only run out a few times a month, topping up Plus is cheaper.

5. A five-developer team

Business billed annually: 5 × $20 × 12 = $1,200 a year. Billed monthly: 5 × $25 = $125 a month, or $1,500 a year. Five Plus subscriptions would be 5 × $20 × 12 = $1,200 a year too, but without SSO, admin controls, larger cloud VMs or shared workspace credits.

Using Codex with an API key

Codex CLI, the Codex SDK and the IDE extension can sign in with an OpenAI API key instead of a ChatGPT account. Usage is then billed at standard OpenAI API pricing: for example GPT-6.1 Sol at $2 input, $0.10 cached input and $10 output per 1M tokens. There is no subscription and no five-hour window, only your API rate limits and prepaid balance. You don't get cloud features such as automatic GitHub code review or Slack.

This works well if you use Codex occasionally, need a model that has been retired from ChatGPT sign-in (the GPT-5.5 retirement on October 14, 2026 did not affect API-key sessions), or want to keep working after hitting plan limits. Any user can run extra local chats with an API key at standard API rates.

Agents can loop. Before running Codex unattended on an API key, set a recharge limit in your OpenAI billing settings so a stuck session can't keep buying credits.

Which option is right for you?

• Trying Codex out: Free or Go with GPT-6 Luna, or a few dollars of API credit.
• Coding a few hours a week: Plus at $20, with credits for the odd busy week.
• Coding with an agent most of the day: Pro 100 or Pro 200.
• Need the lowest latency on Astra or 6.1 Sol: Pro 500, the only personal plan with Ultrafast.
• A team that needs SSO and admin controls: Business.
• Occasional use or CI scripts: an API key.

How Codex compares with other coding tools

Claude Code works the same way, bundled with Claude Pro at $20 and Max at $100 or $200 (see Claude Code pricing). GitHub Copilot starts at $10/month and can hand tasks to Codex as a third-party agent on higher plans (GitHub Copilot pricing). Cursor sells its own tiers with third-party models billed at API prices (Cursor pricing). If you mainly want a free option, our is Codex free guide covers every legitimate route.

Recent changes

• August 31, 2026: GPT-5.4 and GPT-5.4-mini retired from Codex with ChatGPT sign-in, replaced by GPT-6 Sol and GPT-6 Luna.
• September 14, 2026: GPT-5.3-Codex-Spark retired.
• September 29, 2026 (DevDay): GPT-6.1 Sol launched; Pro split into $100, $200 and $500 tiers; Astra Ultrafast came to Codex on Pro 500 and eligible Enterprise and Edu plans.
• October 8, 2026: GPT-6.1 Sol Ultrafast became available in Codex on Pro 500 and eligible Enterprise and Edu plans (off by default for Enterprise until an owner enables it).
• October 14, 2026: GPT-5.5 retired from Codex with ChatGPT sign-in. Paid plans move to GPT-6.1 Sol, and Free and Go to GPT-6 Luna.
• Through at least November 21, 2026: GPT-5.6 Sol is on promotional pricing.

Free and discounted routes

Verified university students in the US and Canada can get $100 in Codex credits; see Codex credits for students. US college students can also claim four free months of ChatGPT Plus until October 31, 2026, which includes full Codex. For more options, see how to get free AI credits.