OpenAI Codex Pricing (2026): Plans, Limits, and Extra Credits
Codex is included in ChatGPT plans: Free US$0/month, Go US$8/month, Plus US$20/month, and Pro from US$100/month, according to the official pricing documentation, checked September 18, 2026. These are the public USD reference rates, not a local-currency or tax-inclusive checkout quote. API-key usage is billed separately.
Table of Contents
- What is OpenAI Codex pricing based on?
- Which ChatGPT plans include Codex?
- How do Codex limits work?
- When should you buy extra credits?
- Codex access versus API billing
- Which plan fits your coding workload?
- Frequently asked questions
What is OpenAI Codex pricing based on?
OpenAI Codex pricing is based on the ChatGPT plan attached to your account, plus optional credits when you need additional usage. Codex is not priced as one universal standalone developer seat with one fixed allowance.
Available usage depends on your ChatGPT plan, whether the task runs locally or in the cloud, the selected model, task size and complexity, weekly or plan-specific limits, and credits purchased after included usage.
Two people on the same plan may complete different numbers of tasks. A small code edit consumes different capacity from a large repository analysis, multi-file refactor, or long-running cloud job.
Think about Codex cost as capacity per workflow, not cost per prompt. Occasional assistance may fit included access. For regular engineering work, ask whether the allowance supports your normal weekly rhythm.
Which ChatGPT plans include Codex?
The official plan table supplies the following USD rates. Enterprise and Edu are sales-led; the Business annual minimum is two users.
| Plan | Published USD price | Billing qualification |
|---|---|---|
| Free | $0/month | Limited included usage |
| Go | $8/month | Monthly reference rate |
| Plus | $20/month | Monthly reference rate |
| Pro | From $100/month | 5x or 20x usage options; starting price is not both options' price |
| Business | $20/user/month annual; $25/user/month monthly | Annual offer requires 2+ users |
| Enterprise / Edu | Quote | Organization-specific terms |
Free and Go are useful for lighter work. Test a representative job before treating either as the capacity budget for a sustained engineering workload; included access is not a guarantee that any size of job will fit.
Plus can suit recurring focused work. For heavier use, check the selected Pro allowance and price together instead of treating every Pro subscription as the same package.
Business and Enterprise include Codex within organizational ChatGPT access. Their suitability depends on the wider workspace and team needs, not an isolated Codex seat.
What would two Business seats cost?
Using the rates above, two annual Business seats cost 2 × US$20 × 12 = US$480/year before tax and extras. Month-to-month seats cost 2 × US$25 = US$50/month, or US$600 over twelve unchanged months. The US$120 difference buys flexibility; annual billing is not a monthly cancellation option. This is arithmetic on published rates, not a purchased account or measured productivity result.
How do Codex limits work?
Codex limits are usage allowances; the number of tasks is not fixed. Task counts vary by model and job size.
A short local task may use less capacity than a cloud job examining a large repository, modifying files, running a broad suite, and requiring iteration. Model choice also affects how much work fits.
Local messages and cloud tasks share usage, and the official documentation also states that ChatGPT Work and Codex share allowance. Moving a task between surfaces does not create a second budget. Source: usage-limit documentation.
Weekly limits may apply. When reached, you may need to wait for refresh or use available credits, depending on account options.
The practical rules are:
- do not estimate capacity only by counting prompts;
- keep large jobs focused;
- expect repository size and complexity to affect consumption;
- treat local and cloud usage as one allowance;
- check current usage before a heavy sprint.
A developer asking for small, well-scoped changes may complete more work than someone submitting broad autonomous tasks. That is normal for workload-based access.
How should you estimate a week of Codex work?
List the jobs you actually expect: small fixes, code review, feature implementation, repository investigation, and cloud tasks. Mark which require a heavier model or broad context. Run a representative week before upgrading and note where limits interrupt useful work.
Do not convert one unusually easy task into a universal task count. Use a range based on your own repository and workflow. The official pricing table is the only current source for model-specific ranges, and those ranges can change.
When should you buy extra credits?
Credits extend usage when you need more capacity than the plan includes. They are most useful for a temporary spike rather than a permanent mismatch.
Credits may make sense when finishing a release, performing a one-time migration, clearing an unusual bug backlog, evaluating a large repository, or completing a short-term project above your weekly rhythm.
If workload consistently exceeds the included allowance, compare your observed extra-credit spend with the price difference for the Pro option you would actually buy. An upgrade is not automatically cheaper merely because its usage multiplier is larger.
Before using credits, identify the cause. Large jobs, heavier models, and repeated retries can consume the allowance. Better task scoping reduces waste, but cannot eliminate the capacity needed for genuinely large work. Credits extend usage; they do not make access unlimited.
How does Codex access differ from API billing?
ChatGPT Codex access and OpenAI API billing are separate paths.
Your subscription covers eligible signed-in Codex usage. Codex CLI, SDK, or IDE workflows can also use an API key, billed separately. Cloud features such as GitHub code review are not included in that API-key route. Source: official pricing.
If you want models programmatically in your application, backend, automation, or developer tool, consult OpenAI API pricing. If you want Codex as an agent in ChatGPT, evaluate ChatGPT plan allowances and credits.
For a concrete API example, GPT-5.3-Codex's official model page lists US$1.75 per million input tokens, US$0.175 cached input, and US$14 output, checked September 18. One million uncached input tokens plus 100,000 output tokens therefore cost US$1.75 + US$1.40 = US$3.15 before other charges. This is a named-model calculation, not a universal Codex task price or a recommendation to choose that model.
Which plan fits your coding workload?
Choose Free or Go to explore Codex or handle lighter work. Use the result to judge fit, rather than assuming the amount of available usage will support every future project.
Choose Plus when your need is recurring but focused. It is designed for weekly coding sessions rather than constant high-volume agent use.
Compare Pro when Codex is a regular engineering tool and your measured workload exceeds Plus. Record interruptions and completed, tested work before paying for more capacity; repository size and repeated retries can matter more than prompt count.
Choose Business or Enterprise when Codex belongs in an organization’s ChatGPT workspace. Consider how many people use coding assistance, how consistently, and which workspace plan is appropriate.
A sensible process is to start with the plan matching current workload, observe how often you hit limits, improve task scoping, use credits for temporary spikes, and move higher only when heavy usage becomes routine.
For adjacent comparisons, see ChatGPT Enterprise pricing, Cursor AI pricing, Manus AI pricing, and the ChatGPT Visibility Checker.
Review a month of completed work, not just sessions opened. A plan is valuable when Codex finishes tested changes with less engineering time, not when it merely generates more activity.
Frequently Asked Questions
How much does Codex cost?
The current public USD reference prices are listed above, with Codex included in eligible ChatGPT plans. Verify regional currency, tax, selected allowance, and any extra credits at checkout.
Does Plus include unlimited Codex usage?
No. Plus supports focused weekly coding sessions, but allowances and weekly limits may apply. Task capacity varies by model and job size.
Does Pro provide more Codex usage than Plus?
Yes. Pro has 5x and 20x usage options relative to Plus. Check the selected option's price and limits; a starting price does not price every variant.
Are Codex credits the same as API credits?
No. Codex credits extend eligible Codex usage within ChatGPT. API usage follows separate API pricing.
Sources: official plan documentation, API pricing, and the named API model's rates.
FAQ
How much does Codex cost?+
Checked September 18, 2026: the official USD plan list shows Free $0/month, Go $8/month, Plus $20/month, and Pro from $100/month. Business starts at $20/user/month billed annually with at least two users, or $25/user/month billed monthly.
Does Plus include unlimited Codex usage?+
No. Model choice, task size, shared usage, and account limits affect how much work fits in the included allowance.
Does Pro provide more Codex usage than Plus?+
Yes. Official documentation lists Pro options with 5x or 20x the Plus usage limits. Confirm which option you are purchasing; the advertised starting price is not a quote for both.
Are Codex credits the same as API credits?+
No. Credits extend eligible subscription usage. API-key work follows separate API billing and does not consume a ChatGPT subscription's included allowance.
Sources
Is AI recommending you?
Run a free AI visibility audit and find out in under a minute.