Codex Pricing, OpenAI Codex Pricing and ChatGPT Codex Plans Compared

Codex costs $20 a month on the cheapest plan that includes the full product, ChatGPT Plus. Heavy individual users pay $100 for Pro 5x, and companies pay $20 per ChatGPT Business seat billed annually or $25 billed monthly, with a $100 seat for heavy users. Free and Go ($8 a month) include only GPT-6 Luna in the desktop app. Prices exclude sales tax.

What those prices buy is a share of messages, not a fixed number of them. OpenAI's pricing doc says the count "depends on the model used, size and complexity of your tasks, and whether you run them locally or in the cloud," and it publishes its message figures as estimates per five-hour window with weekly limits that "may also apply." It also notes that ChatGPT Work and Codex share one allowance, so a developer who also uses ChatGPT Work for research is spending the same pool on both jobs.

The unit that stays constant is the multiple. Pro is 5x or 20x Plus, a Business Standard seat carries the same estimates as Plus, and OpenAI says the $100 Business seat "uses the Pro 5x estimates." Divide each price by its multiple and you get the price of one Plus-sized allowance on every plan:

OpenAI charges $20 per Plus-sized allowance almost everywhere, which means moving a developer from Plus to Pro 5x buys capacity, not a discount. The one plan that halved the unit price, Pro 20x at $10 per allowance, is the one OpenAI stopped selling to new subscribers on September 10, 2026. Existing Pro $200 subscriptions keep working, and Pro $100 is not affected. The monthly Business seat is the most expensive capacity on the list, so a team that expects to keep Codex for a year should take annual billing.

Codex usage limits are counted in local messages per five-hour window, and the count depends on the model. On Plus, OpenAI estimates 15 to 150 GPT-6 Sol messages, 5 to 45 on GPT-6 Astra and 350 to 3,000 on GPT-6 Luna per window. Pro multiplies those by 5 or 20. Weekly limits can apply on top, and cloud chats use more of the allowance than local ones.

The spread inside each cell is tenfold because a one-line fix and a long refactor that holds a large repository in context are both "a message." OpenAI's own guidance is that model choice, context, reasoning, tool use and caching all move consumption, and that "prompt length alone isn't a reliable estimate." The practical reading is the ratio between rows: on the same seat, GPT-6 Sol gives you about three times as many messages as GPT-6 Astra, and GPT-6 Luna about twenty times as many as Sol.

Three details change the math in practice. Cloud chats on ChatGPT plans run on GPT-5.6 Sol, which gets fewer messages per window than GPT-6 Sol, so cloud-heavy teams burn allowance faster. Image generation uses limits three to five times faster than a text turn. And GPT-5.5 retires from ChatGPT, ChatGPT Work and Codex on all plans on October 14, 2026, so anyone who pinned it as a default needs a new one. Enterprise and Edu workspaces on flexible pricing have no fixed rate limits; usage simply draws down credits.

A Codex credit costs $0.04. OpenAI never prints that number on the credit rate card, but its two rate cards divide to it exactly: GPT-6 Sol is 50 and 250 credits per million input and output tokens on the credit card and $2.00 and $10.00 on the Enterprise dollar card. Every model row lands on four cents, so any credit figure OpenAI publishes converts straight into a budget line.

GPT-6 Sol, added to Codex on September 22, 2026, costs exactly half of GPT-5.6 Sol per token. OpenAI says a typical GPT-5.6 Sol task uses 5 to 30 credits, which is $0.20 to $1.20. The same token volume on GPT-6 Sol would come to roughly $0.10 to $0.60; that is our arithmetic from the published rates, not an OpenAI figure, and real tasks vary with how many tokens each model spends. OpenAI's one aggregate number is that Codex costs "approximately $100 to $200 per developer per month" on average, or 2,500 to 5,000 credits.

The multipliers are where bills surprise people. Fast mode uses 2.5 times the Standard credit rate on GPT-6 Astra, Sol and Luna, so a $1.00 task becomes $2.50 because someone preferred quicker answers. Voice in the desktop app costs $0.05 a minute (1.25 credits on credit-billed workspaces) plus the tokens of the model doing the work. On the plus side, Codex credit billing has no separate cache-write charge. Personal-plan credits are valid for 12 months from purchase, are non-refundable except for billing errors or unauthorized use, and can be bought once a plan limit is reached from Settings, then Usage, in ChatGPT or Usage and Billing in the Codex app.

For a company, Codex comes through ChatGPT Business or ChatGPT Enterprise. Business is self-serve at $20 per user per month billed annually or $25 monthly, for two or more users, with a $100 seat for heavy users that carries Pro 5x limits. Enterprise and Edu are quoted by sales and can run on flexible pricing, where there is no fixed per-seat limit and all usage draws down credits.

What the extra money buys on Enterprise is mostly control, not Codex capability. OpenAI's feature table gives Business SAML SSO, MFA, workspace user management, cloud-managed config policies and no training on business data by default. Enterprise adds RBAC and custom roles, SCIM, EKM and domain verification, retention and residency controls, the analytics dashboard and API, the Compliance API with audit logs, and Codex Security for connected GitHub repositories. Eligible Business workspaces and Enterprise can also assign usage-based Codex-only seats, which include Codex only and draw on workspace credits from the first task instead of a seat allowance.

Here is what a 10-developer US team pays at list, before any credits:

The last row is the one to plan around. OpenAI's own average says a working developer consumes five to ten times what a $20 seat costs, and that consumption has to be paid for either with bigger seats or with credits. Seats are the cheaper way to buy predictable daily use; credits are the right way to cover the occasional heavy week. A team that gives its three heaviest users $100 seats and keeps everyone else on Standard, with a shared credit pool and a spend cap for spikes, usually lands well under the all-credit figure. Admins can see usage in Codex settings, and on Enterprise in the global admin console.

Codex can run on an OpenAI API key instead of a ChatGPT login, billed at API token rates with no seat fee. On OpenAI's API price list GPT-6 Sol costs $2.00 per million input tokens, $0.20 cached and $10.00 output, the same headline rate as a Codex credit. The API path covers the CLI, the SDK and the IDE extension, but none of the cloud features.

The headline rates match; the fine print does not. OpenAI's API price list charges $2.50 per million tokens for cache writes on GPT-6 Sol, while Codex credit billing has "no separate cache-write charge." Long context on the API costs $4.00 in and $15.00 out per million tokens for GPT-6 Sol. Fast mode is listed at $4.00 and $20.00 on the API, twice the Standard rate, against 2.5 times on the credit card. And an API key loses Codex cloud, GitHub code review with @codex, the Slack and Linear integrations, and ChatGPT Voice.

The break-even is simple. A $20 Plus plan equals 500 credits of API-rate usage, which at OpenAI's own GPT-5.6 Sol estimate of 5 to 30 credits per task is somewhere between 17 and 100 tasks a month. Anyone doing more than that should be on a plan, and anyone using Codex daily should not be paying API rates at all. OpenAI itself positions the API key for "automation in shared environments like CI," and every user can fall back on a key for extra local chats once a plan limit is hit.

Codex, Claude Code, Cursor and GitHub Copilot all start near $20 a month for one developer, and all four now meter heavy use by tokens or credits on top of the seat. The differences are in the heavy tier and the team seat, where Codex and Claude Code both charge $100 for five times the base usage.

For an individual who needs the most capacity, note that Codex's $200 tier is closed to new buyers, while Claude Max 20x and Cursor Ultra, both at $200, are still on sale. For a team, Codex Business and Claude Team Standard cost the same at $20 annual, and both put a $100 seat above it for heavy users. Our breakdowns cover each one in detail: Claude Code pricing, Cursor pricing, GitHub Copilot Enterprise pricing and Devin pricing for the autonomous-agent end of the market.