ChatGPT and Codex plans, pricing and usage limits

I want the budget conversation to start with the job being paid for. A person drafting campaign copy, a builder fixing a website and an API service scoring leads can all be described as “using OpenAI.” They do not create the same bill.

I call this the three-line budget: seats, additional usage and software requests. Put them on one page before deciding a plan is cheap. Otherwise the expense report becomes a scavenger hunt.

Start with three different bills

The seat is the recurring product subscription. Additional usage can extend work beyond an included allowance where supported. API requests are metered under an API organization and project. A workflow may use more than one of these, so an invoice total without the route is difficult to interpret.

For fictional Cedar Metrics, account executives might work in ChatGPT while an operations builder uses Codex and a small service calls the API. The company should be able to identify which budget pays for each activity. That is ordinary ownership, not an elaborate finance system.

I would keep a row for every recurring workflow: owner, product, sign-in method, expected cadence and billing account. A nightly task without an owner is a subscription to future confusion. The first time the bill rises, this row tells you where to look.

What the published prices establish

As checked September 22, 2026, the official Work and Codex pricing page lists Plus at $20 per month and Pro from $100, with higher usage tiers. Standard Business is listed at $20 per user per month billed annually or $25 monthly, for at least two users. Enterprise and Edu require the applicable commercial arrangement.

These figures establish a starting comparison, not every regional price, promotion or entitlement. Check your checkout and workspace offer before submitting a purchase request. Free and Go access, included capacity and model availability can vary; the visible account allowance is more useful than copying another person's message count.

For a beginner, the right question is which required feature or interruption an upgrade would solve. A larger plan that does not change that constraint is simply a larger bill. Keep the required capability next to the price so the decision remains concrete.

Understand shared Work and Codex usage

The current documentation says Work and Codex share usage. A long research task and a development session can therefore draw on the same available capacity. Counting messages alone misses how much context, reasoning, tool use and execution sits behind each one.

A short prompt asking for an entire campaign workbook may consume more than several focused revisions to one paragraph. That is why “I only sent five prompts” is a poor spending explanation. The output contract and the work required to satisfy it are the more useful unit.

Review the usage dashboard and the session's reported limits. If important work repeatedly stalls, first inspect whether tasks are unnecessarily broad or repeatedly rereading the same materials. Then compare a larger allowance or a different model. An upgrade should solve observed friction, not anxiety about running out someday.

Choose personal capacity by observed work

A personal plan can be enough when one person needs the product for individual work and the relevant data policy permits it. Higher usage may justify Pro when valuable sessions repeatedly exhaust a lower allowance. The official page describes 5x and 20x Pro tiers, but those are relative usage levels, not a universal number of guaranteed tasks.

I would run a normal week and note interruptions, task types and accepted outputs. If most requests are short drafts, the decision may look very different from someone running long repository changes all day. The point is to measure your workload instead of borrowing someone else's screenshot of a usage meter.

Keep API automation out of this comparison unless it actually uses the API. Supported app tasks can run unattended too. A schedule identifies when the work happens; it does not identify how it is billed. Confirm the product running the task before calculating the cost.

Buy team plans for the operating requirements

A team decision includes administration, identity, sharing and data controls. Business and Enterprise should be compared on the requirements your organization actually has, alongside usage. SSO, member management and approved connected services matter even when nobody approaches a usage limit.

For example, twelve standard Business seats at the listed annual rate represent a $240 monthly equivalent under an annual commitment. At the listed monthly rate they would be $300 per month. This is simple illustrative seat arithmetic, excluding taxes, additional usage and API charges. Put those exclusions beside the number.

Different workspace arrangements and seat types can change the comparison. Have the workspace owner confirm access to the required features and models before rollout. Purchasing a seat is only one part of making a workflow available; plugin policies, source-system permissions and local controls still apply.

Treat API-key sign-in as a separate route

Codex can use supported API-key authentication routes, where usage follows API billing and model access. That is distinct from signing in with a ChatGPT plan. Some cloud-based product features are not available through API-key access, so it is not merely the same subscription with a different payment method.

If you build a service, create the appropriate API project and keep its credentials server-side. Set an owner and monitor usage at that project. A subscription receipt from ChatGPT does not prove that the API project has budget or access to a particular model.

For forecasting, estimate representative requests and include tools, reasoning, retries and validation. A rate per token is an input to the calculation, not the total cost of operating the service. The API guide starts with a deliberately small request and a visible output.

Make a budget that survives real use

Use a simple ledger with seats, additional usage, API requests and review effort. Then divide by accepted results for the workflow you care about. A monthly total becomes more useful when you can say which reports, briefs or repository changes it produced.

For a weekly account brief, count completed briefs and the time someone spends verifying them. For a content pipeline, count publishable pieces and the effort needed to correct unsupported claims. A larger model may lower review cost; a shorter output may lower both usage and editing cost. These are hypotheses to test, not promised savings.

Set a review cadence before launch. Inspect unexpected spikes, abandoned experiments and tasks running more often than intended. You do not need to audit every token to find a daily job that was meant to be weekly. A named owner and a visible schedule do a surprising amount of financial work.

Where pricing comparisons mislead

The first mistake is treating an estimated message range as a contract. Limits depend on the model, complexity and environment, and account dashboards can change. The second is comparing annual monthly equivalents with monthly billing without showing the commitment. Both errors make an attractive spreadsheet that disappoints at checkout.

Another trap is assuming shared usage means identical feature access. A model or tool can remain unavailable because of plan eligibility, workspace policy or rollout. Increasing spend does not automatically clear every access boundary. Diagnose the missing capability before changing the budget.

Finally, avoid claiming a plan is worth it before naming the result. I would judge a month by useful work completed and review effort, with the actual bill beside it. The team controls guide covers the access side. Which part of your current work would justify the first paid month?

How to set it up

Frequently asked questions

Sources & further reading

• Official Work and Codex pricing
• Authentication routes
• Workspace usage controls
• Roles and product access

ChatGPT and Codex change quickly. This page was last reviewed September 22, 2026; verify time-sensitive details against the official docs above before relying on them.