Codex Pricing: Plans, Credits and API Costs

Published on October 31, 2026 by Steven Jones

Codex pricing has three distinct routes: use the allowance included with a ChatGPT plan, buy credits when eligible included usage runs out, or sign in with an API key and pay separately for tokens. They are not interchangeable. A $20 subscription does not include a $20 API balance, and a credit estimate does not automatically tell you what a coding task costs in dollars.

This guide uses OpenAI’s published rates checked on 31 October 2026, alongside explicit token and caching assumptions. It covers which interface each plan supports, why usage limits sometimes disappear faster than expected and how to budget repeatable coding work. Verify changes against the official Codex and ChatGPT Work pricing documentation before purchasing credits or changing plans.

Codex pricing at a glance: which route should you pay for?

Route	Published USD price	What it covers	Best fit
ChatGPT Free	$0/month	Limited local Codex access in the desktop app, subject to rollout; no Codex Cloud	Trying a short coding task
ChatGPT Go	$8/month	Lightweight local use; no Codex Cloud	Occasional small changes
ChatGPT Plus	$20/month	Included Codex usage across supported desktop, CLI, IDE and cloud interfaces; limits apply	A few focused coding sessions each week
ChatGPT Pro	$100, $200 or $500/month	Larger included allowances; Pro 500 offers eligible Ultrafast access	Frequent or lengthy agent sessions
ChatGPT Business	$20/user/month annually or $25/user/month monthly; minimum two users	Team workspace controls and included usage, with credit-based extension options	Small development teams
Enterprise / Edu	Contract pricing	Workspace-specific controls and credit-based or USD token-based arrangements	Governed organisation-wide use
API key	Usage-based, no Codex subscription required	Separate model/token billing in the CLI, IDE or SDK; no subscription-only Codex Cloud features	CI jobs, scripts and measurable automation

Prices are US list prices, not guaranteed UK checkout totals. Taxes, currency conversion, regional availability and workspace agreements can change the amount paid. The practical starting point is Plus for interactive coding and API billing for repeatable automation, unless the work actually requires team governance or substantially higher included usage.

---

---

Included Codex usage is an allowance, not a token wallet

With ChatGPT sign-in, Codex uses the included allowance attached to your plan. It does not deposit the credit equivalent of your subscription fee into an API account. Codex and ChatGPT Work share usage on supported plans, so heavy work in one interface can reduce what remains in the other. Other eligible agentic features may also share the same allowance.

OpenAI’s October 2026 guidance estimates roughly 15 to 160 local GPT-6.1 Sol messages within a five-hour window on Plus. GPT-6 Luna has a much higher estimated range, while GPT-6 Astra has a lower one. These are illustrations, not guaranteed message quotas. Two prompts can consume radically different amounts if one repeatedly scans a repository, runs tests, and delegates work.

Plus and Standard Business may have both five-hour and weekly limits. When both apply, you need headroom in both. Pro 100, Pro 200, and Pro 500 currently have no five-hour limit in Work and Codex, but still include allowances and weekly or model-specific restrictions. A higher-tier subscription is not unlimited coding.

The model, reasoning setting, running tools and the size of your working context affect allowance use. For example, an oversized AGENTS.md and unnecessary MCP tool definitions can repeatedly enlarge the material the agent processes. Our Codex CLI MCP setup guide shows how to restrict those integrations rather than leaving everything enabled.

What happens when you reach a five-hour or weekly limit?

First, inspect Settings > Usage or the Codex usage dashboard. From an active CLI session, run /status. Identify the actual restriction: five-hour allowance, weekly allowance, a separate model limit, or an exhausted purchased-credit balance. Don’t buy credits assuming every restriction is bypassable.

For an eligible account, the options shown may include waiting for the displayed reset, switching to a less costly model, adding credits, upgrading or purchasing an immediate usage reset. Purchased credits extend eligible usage after you consume included allowances. A paid reset is a different product: on eligible Plus and Pro accounts, it brings the weekly allowance forward and starts a new weekly period with the first subsequent Work or Codex request. It does not permanently increase the plan.

OpenAI says a five-hour window begins with your first eligible request after the preceding window ends; the weekly allowance resets on its own schedule. Check the account’s displayed times rather than assuming both reset together. If a task hits a limit partway through a turn, it may be allowed to finish, subject to fair-use controls, but that does not guarantee further turns.

Developers repeatedly report surprisingly rapid allowance depletion during multi-step coding sessions. A sensible response is to log the model, speed and task scope before buying a bigger plan. If every large change launches several agents and repeatedly reads the same files, increasing the subscription may only hide an expensive workflow.

Codex credit rates: the current token table

Rate check: 31 October 2026. The following official rates are for Standard speed and are measured in credits per one million tokens. They apply to eligible credit-metered use, not as a conversion of your included subscription allowance into credits.

Model	Uncached input	Cached input	Output
GPT-6 Astra	250	25	1,250
GPT-6.1 Sol	50	2.5	250
GPT-6 Sol	50	5	250
GPT-6 Luna	2.5	0.25	12.5
GPT-5.6 Sol	100	10	500
GPT-5.6 Terra	50	5	300
GPT-5.6 Luna	5	0.5	30

Use this formula, counting uncached input and cached input separately:

Credits = (uncached input tokens x input rate + cached input tokens x cached rate + output tokens x output rate) / 1,000,000

Codex credit billing has no additional cache-write line item. API-key billing may have different caching charges. For purchased credits, supported Fast mode consumes 2x the Standard credit amount; it consumes included subscription allowance at 2.5x. Eligible Ultrafast modes have still higher multipliers. Credit purchase prices and discounts vary by plan or agreement, so a 20-credit task is not necessarily a fixed-dollar purchase.

Three editable Codex budget examples

These are worked assumptions, not measured coding tasks. You can change the token counts, task count or model in each WordPress table and recalculate using the formula above. All examples use Standard speed, with cache hits specified explicitly.

Example 1: an outdated GPT-5.6 Sol table changes the answer

Editable assumption	Value
Billing route	Purchased Codex credits
Model / mode	GPT-5.6 Sol / Standard
Uncached input tokens	100,000
Cached input tokens	40,000
Output tokens	20,000
Current rate calculation	(100,000 x 100 + 40,000 x 10 + 20,000 x 500) / 1,000,000
Current estimate	20.4 credits

Some older pricing tables show 125 input, 12.5 cached-input and 750 output credits per million tokens for the same model. On these exact assumptions, they produce 28 credits, overstating the current calculated usage by 7.6 credits. The current 100/10/500 GPT-5.6 Sol rate is published as promotional pricing available at least until 21 November 2026. Recheck it after that date.

Example 2: buying credits for 50 additional Codex tasks

Editable assumption	Value
Billing route	Purchased Codex credits after included usage
Model / mode	GPT-6.1 Sol / Standard
Additional tasks	50
Tokens per task	100,000 uncached input; 40,000 cached input; 20,000 output
Credits per task	(100,000 x 50 + 40,000 x 2.5 + 20,000 x 250) / 1,000,000 = 10.1
50-task estimate	505 credits
Fast-mode equivalent, if supported	1,010 purchased credits at 2x

To find the cash cost, multiply 505 by the effective price per purchased credit actually shown in your account. Do not assume every one of those 50 tasks would consume 10.1 credits: repeated failed runs, output length and caching can change the bill. This helps you decide whether an occasional top-up makes sense or whether a larger Pro allowance deserves testing.

Example 3: a separately billed API workflow

Editable assumption	Value
Billing route	API key, charged in USD
Model / processing	gpt-6.1-sol / Standard, short-context rates
Monthly jobs	300
Tokens per job	30,000 uncached input; 70,000 cached input; 10,000 output
API rate per 1M tokens	$2.00 input; $0.10 cached input; $10.00 output
Cost per job	(30,000 x $2 + 70,000 x $0.10 + 10,000 x $10) / 1,000,000 = $0.167
300-job estimate	$50.10

This assumes the stated 70,000 tokens really qualify as cached input. If none of the 100,000 input tokens is cached, the same job rises to $0.30, or $90 for 300 jobs. Neither calculation includes any applicable API cache-write charges, container sessions, extra tools, tax or regional processing uplift. Record those separately rather than hiding them inside a supposed flat cost per task.

Your search, your sources

Make DIY AI a preferred source

See more of our reporting in Google Top Stories, AI Overviews and AI Mode.

Add DIY AI on Google

When an API key makes more sense than upgrading Pro

Choose API billing when a job needs to run unattended in CI, an SDK or a script, and you can track its token usage. That provides a measurable cost for each execution and avoids buying a larger personal subscription purely to support automation. The trade-off is operational: an API key needs its own budget monitoring, access control and management. API usage doesn’t inherit your Plus or Pro included allowance, and an API key doesn’t unlock Codex Cloud’s subscription-only integrations.

For interactive editing in a local repository, Plus is usually the cheaper experiment because the subscription includes other ChatGPT benefits. Move to Pro only after recording several days of actual limit interruptions. A $100 or $200 plan is a weak upgrade if the root cause is unnecessary context or a task that should have been split into smaller verified steps.

Compare competing agents on the same accepted change, not the number of prompts. Our Claude Code pricing breakdown helps with the alternative billing model, while Claude Code vs OpenAI Codex covers the wider workflow decision. Developers choosing an extensible multi-provider setup can also read OpenCode vs Codex.

How to prevent a misleading Codex cost estimate

• Confirm the account route. ChatGPT sign-in spends included usage or eligible credits; API-key sign-in bills the API account separately.
• Separate cached from uncached tokens. Never count the same cached input twice, and do not budget on an assumed cache hit before measuring it.
• Log the speed setting. Standard, Fast and Ultrafast can consume very different allowances or purchased credits for the same token workload.
• Inspect repeat work. Large instructions, too many MCP servers, long tool output and retries can make a short-looking instruction costly.
• Review results, not just spend. Track cost per accepted change after tests and review. A cheap failure followed by several retries is not a cheap task.

A repeatable workflow also benefits from smaller, documented instructions. Our Codex skills guide explains how to package instructions so you can check them rather than reinvent them in every session.

Codex pricing FAQs

Is Codex free?

Yes, limited Codex access is included on Free, subject to supported clients and rollout. That does not include the broader Codex Cloud capability offered on eligible paid plans, and it is not a free API allowance.

Do ChatGPT Plus or Pro subscriptions include API credits?

No general API allowance is included merely because you subscribe. Credits bought to extend eligible Codex or Work usage are a separate mechanism from the API platform’s dollar-based billing.

Do Codex limits reset every five hours?

Not all of them. Some plans have five-hour windows and weekly allowances. Pro tiers currently have no five-hour Work and Codex limit, but weekly and other limits can still apply. The reset time in your usage dashboard is more reliable than a generic timetable.

Which Codex plan should I start with?

Start with Free or Go for supported lightweight local tasks, Plus for regular human-guided coding, and the API for measurable automation. Upgrade to Pro when actual session history shows that the included allowance, rather than poor task scoping, is your bottleneck. Recalculate the credit or API bill whenever the model, cache mix or processing mode changes.