ChatGPT Enterprise Pricing: ChatGPT Business Pricing, ChatGPT Team Pricing and OpenAI Enterprise Cost per Seat

OpenAI does not publish a ChatGPT Enterprise seat price. Its business pricing page shows Business at $20 and $100 a seat and, next to Enterprise, the words "Custom pricing" with a contact sales button, plus a short list of what the contract adds: volume discounts, invoicing and billing, SLAs and custom legal terms. The help center describes the same thing as a process: Enterprise "is purchased at the organization level," and the way in is a sales form.

So the honest answer has two halves. The seat half is negotiated and not public. The usage half is public in more detail than almost any other AI vendor publishes, and it is the half that grows. OpenAI maintains two rate cards for it, one in credits and one in dollars per million tokens, and as the next sections show, the two agree once you know what a credit costs.

A figure of about $60 per seat per month with a 150 seat minimum circulates widely in third-party pricing guides. We could not find either number on any OpenAI pricing page or help article on September 10, 2026, so treat it as a reported negotiating anchor rather than a list price. What OpenAI does state is that Business stops at 200 seats for new workspaces, which means above 200 people Enterprise is the only plan you can buy.

ChatGPT Business is the self-serve plan and its prices are fully published. A Standard seat is $20 per user per month billed annually, or $25 billed monthly. A Premium seat is $100 per user per month billed annually, or $125 monthly, and carries 5x the usage of a Standard seat with no 5-hour usage limit. You can mix both types in one workspace and move people between them, so the usual pattern is Standard for most of the team and Premium for the handful who live in ChatGPT Work or Codex all day.

If you are searching for ChatGPT Team pricing, this is it. OpenAI renamed ChatGPT Team to ChatGPT Business on August 29, 2025, and says the rename itself changed nothing. The price did change later: on April 2, 2026 OpenAI cut subscription seats by $5 a month, which is how Standard reached today's $20 and $25. Premium seats became generally available on August 25, 2026.

The plan rules that matter for a budget:

The payment row is the one that quietly pushes companies toward Enterprise. If your finance team pays software on invoice against a purchase order, Business is not an option at any seat count, and OpenAI's help center says support cannot convert a self-serve Business workspace to invoice billing.

OpenAI prices everything above a seat's included usage in credits and never prints the dollar value of a credit on its rate card. It did print it somewhere else. When it launched Premium seats for Business, it offered early buyers "$100 worth of workspace credits (2,500 credits)" per Premium seat. That is $0.04 per credit.

The second rate card confirms it independently. New Enterprise customers billed in US dollars get a token rate card priced in dollars per million tokens, while everyone else gets the same models priced in credits per million tokens. Divide one by the other:

Every row lands on exactly four cents, cached input included (GPT-5.6 Sol cached input is 10 credits on one card and $0.40 on the other). The only visible gap is a rounding one: GPT-5.4 mini output is $4.50 on the dollar card, which is 112.5 credits, and the credit card prints 113. Voice lines up too, at 1.25 credits per minute on the credit card and $0.05 per minute on the dollar card.

One caveat on the GPT-5.6 Sol row: both cards label Sol's Work and Codex rates as promotional pricing "available at least through November 21, 2026," so that row may reprice after the promotion. The other five rows carry no promotion and give the same $0.04, so the credit value does not depend on it.

Two practical conclusions. First, moving from credit billing to dollar billing does not change your unit price at list; it changes the unit your invoice is written in. Any real discount lives in the contract, not in the choice of rate card. Second, once you know a credit is four cents, every credit figure OpenAI publishes becomes a dollar figure you can put in a budget.

These are OpenAI's published credit rates converted at $0.04 per credit. On Business they only apply once a person has used up what their seat includes; on Enterprise they draw down the shared pool. Instant chat is unlimited on both plans, and a message ChatGPT routes to a mini model costs 0 credits.

OpenAI also publishes one aggregate: Codex "costs approximately $100 to $200 per developer per month" on average. At four cents a credit that is 2,500 to 5,000 credits a developer, which puts a heavy Codex user well above the price of the seat they sit in. Reasoning effort does not move the chat rate: Medium, High and Extra High all run GPT-5.6 Sol at the same 10 credits.

Read the table against the seat prices and one comparison stands out. A $20 Standard seat costs the same as 500 credits, which buys 50 GPT-5.6 Sol messages or about 16 agent mode messages as overage. The seat is cheap because the included usage is what you are paying for. Credits are where heavy users become expensive.

OpenAI describes the Premium seat as a multiple of Standard rather than in tokens: 5x the usage. That makes the division easy.

Both seat types land on $20.00 per Standard-equivalent unit on annual billing and $25.00 on monthly billing. There is no volume discount inside Business: a Premium seat is exactly five times the price for exactly five times the usage. The only discount OpenAI sells on the plan is annual commitment, worth 20% on both seat types.

What Premium buys that the arithmetic hides is the removal of the 5-hour limit, plus a larger Pro model allowance of 50 messages a week against Standard's 15 a month. So the seat decision is about interruptions, not cost efficiency. Give Premium to the people who hit the 5-hour wall mid-task and keep everyone else on Standard.

Anthropic prices Claude Team on the identical ladder, $20 and $100 annual, $25 and $125 monthly, also at exactly 5x between the two seats, which we break down in Claude Enterprise pricing. The two biggest model vendors have converged on the same four seat prices.

Enterprise does not change the price of a credit. It changes who holds the credits and how the bill catches up with headcount.

The pool is bought at the contract level. Enterprise and Edu workspaces "purchase a shared credit pool at the contract level," and every user and seat type draws from it. There are no per-seat usage caps by default. Owners and admins can set spend controls by group through role-based access control, and when the pool is exhausted, advanced features pause unless owners have enabled overages or bought more credits through their OpenAI account team.

Overage can be capped at zero. Enterprise owners can configure usage alerts and hard overage limits in workspace billing settings. OpenAI's own example: set the overage limit to 0 and the organization cannot go into overage at all. If your procurement policy needs a hard ceiling, this is the setting to confirm on day one.

Added seats are reconciled quarterly. Adding users does not immediately grow the pool. On standard seat-bundled contracts, qualifying added seats are reconciled through true-ups "typically every 3 months from the workspace's subscription or service-term start date, not calendar quarters," and the matching pooled credits arrive after the true-up. A company that hires 40 people the month after a true-up runs them on the existing pool until the next one.

Codex seats have no fixed price. Enterprise workspaces can assign Codex-only seats, which OpenAI describes as usage-based "with no fixed cost per user per month." Those users cannot open ChatGPT, and every Codex task they run is paid from workspace credits.

Credit expiry is in the Order Form. On Business, credits are valid for 12 months after purchase. On Enterprise, OpenAI says allocation and expiration "are defined in your Order Form," which makes expiry a negotiable term and one worth reading before you sign.

The dollar rate card for new usage-billed Enterprise contracts lists four charges that sit on top of the per-token rate.

Fast mode is the one to watch. A GPT-5.6 Sol task that costs $1.00 at standard speed costs $2.50 in fast mode, and developers switch it on because it feels better, not because anyone approved it. Long context works the same way on large inputs: a 500,000 token input on GPT-5.6 Sol bills at $8.00 per million instead of $4.00, so that one request's input costs $4.00 rather than $2.00. OpenAI makes one exception: GPT-6 Astra in Codex does not incur the long-context multiplier.

US buyers with residency requirements should price the 1.1x before signing. It is a multiplier on all usage, not a feature fee, so $20,000 a month of tokens becomes $22,000, or $24,000 a year attributable to where the processing happens. Anthropic charges the same 1.1x for US-only inference on Claude, and both vendors charge $10 per 1,000 web searches, so neither line separates them.

OpenAI's own comparison table shows the two plans identical on most features, including ChatGPT Work, Codex, workspace agents, company knowledge, apps that connect to internal tools, and the Excel, PowerPoint and Google Sheets extensions. Neither plan trains on your business data by default. The differences sit in context size, administration and support:

OpenAI's Business help article adds three more things that require a contracted plan: Zero Data Retention, business associate agreements (BAAs) and sales-led payment terms. If a security questionnaire has a line for SCIM deprovisioning or a BAA, Business cannot pass it at any price. We walk through the choice seat band by seat band in ChatGPT Business vs Enterprise pricing.

Anthropic and OpenAI are the two vendors most US companies now shortlist for a company-wide assistant, and their published pricing overlaps more than the marketing suggests.

Three things decide it on price. The team tiers are a draw: identical seat prices, the same 5x gap between Standard and Premium, and OpenAI allows 50 more seats before forcing the upgrade. At the enterprise tier Anthropic is more transparent, with a public $20 seat and a self-serve path from 20 seats, while OpenAI's seat line only exists after a sales conversation. And on usage, OpenAI's top model lists at exactly twice Anthropic's on both input and output, and GPT-5.6 Sol at $4 and $20 is likewise twice Claude Sonnet 5 at $2 and $10, although Sol's rate is promotional through at least November 21, 2026. List rate is not cost per task, because models spend different numbers of tokens on the same job, but it is the right starting point for a budget model.

Every price on this page buys access for people. That is the right unit when you are handing a capable assistant to staff who will each use it differently, and ChatGPT Business at $20 a seat is one of the cheapest ways to do that on the market.

It is the wrong unit for one repeatable job. Reading invoices, triaging a support queue, qualifying inbound leads and reconciling statements do not need a person at a keyboard, so a seat license plus a credit pool is an expensive way to buy them: you pay for a seat to administer the work and then four cents a credit for the volume, with agent mode at $1.20 a message. A ready-made agent sold for that one job is priced on the job instead.

So split the evaluation. Price ChatGPT for the people who will use it every day, and price the repeatable work separately. Our comparison of AI agent platforms lays out published seat and consumption rates across the major vendors, how AI agent pricing works covers the three metering models you will meet, and the agents above are priced per job, so you can hold one directly against what the same work would cost in ChatGPT credits.