ChatGPT Enterprise pricing: ChatGPT Business pricing, team plan cost per seat and the 200 seat cap

OpenAI does not publish a ChatGPT Enterprise price. There is no figure on chatgpt.com/pricing, no tier card, and no calculator. The plan is sold by quote through OpenAI sales, billed annually, and paid by invoice rather than by card. Any specific per seat number you read for ChatGPT Enterprise came from a buyer, a reseller or an estimate, not from OpenAI.

The pricing page is direct about it. The plan FAQ reads, verbatim: "Go, Free, and Plus plans are designed to be used by individuals. Business and Enterprise are for businesses. Business plans are available starting at 2 users. Please contact sales to learn more about our Enterprise plan." On payment methods it adds: "You can purchase ChatGPT Go, Plus, Pro, or Business with any major credit card. For ChatGPT Enterprise, please contact sales for alternative payment options such as invoicing."

What OpenAI does publish about Enterprise is the shape of the contract, and that is more useful than a headline rate. Enterprise workspaces buy a shared credit pool at the contract level. All users and all seat types draw from that one pool. There are, in OpenAI's words, "no per-seat usage caps by default", and owners use role-based access control to set spend limits by group instead. Adding people mid-term does not immediately add credits: qualifying seats are "generally reconciled through quarterly true-ups, typically every 3 months from the workspace's subscription or service-term start date, not calendar quarters."

So the honest answer to what Enterprise costs is that it has two moving parts, a negotiated seat count and a negotiated credit pool, and only one of them behaves like a subscription. Budget for both, and ask your account team what happens at the true-up before you sign, not after.

ChatGPT Business costs $20 per user per month on a Standard seat when billed annually, or $25 per user per month billed monthly. A Premium seat costs $100 per user per month billed annually, or $125 monthly. There is a two seat minimum per workspace and you can mix the two seat types freely.

This is the table OpenAI publishes in its Business billing article, restated here without changes:

Premium is the newer of the two. OpenAI describes it as "5x more usage than Standard seats, no 5-hour usage limit, and the flexibility to mix, assign, and reassign seat types." The two seat minimum can be met with one of each, which matters if you have a single heavy user and everyone else on light duty.

Two billing details are easy to miss and both cost money. First: "Unused ChatGPT seats are not refundable once you commit to a seat count on either a monthly or annual plan." Billing follows the billable seat count, not whether anyone signed in. Second, on annual plans, "added seats remain committed through the end of the current annual term. Seat reductions take effect at the start of the next annual billing cycle." Mid-year additions are prorated for the remaining days and then locked in.

OpenAI works the proration example through in the same article. Eight Standard seats at $20 and three Premium at $100 on an annual plan is $1,920 plus $3,600, an initial annual charge of $5,520. Adding two of each on day 100 costs 265 remaining days divided by 365, times $2,880, so $2,090.96 that day, and the annual charge becomes $8,400 from the next cycle.

Two hundred. Since August 24, 2026, OpenAI caps a ChatGPT Business subscription at 200 paid Standard and Premium seats in total, and states that its support team cannot raise the limit for you. Below that ceiling Business is self-serve on a card. Above it, the only route is an Enterprise contract that carries no published price.

The wording in OpenAI's Business billing article is unusually final: "Starting on August 24, 2026, the upper limit is 200 paid Standard and Premium seats total per ChatGPT Business subscription. Our support team is unable to change your maximum seat limit." Most vendor limits come with an escalation path. This one is written to close it off.

This is the same shape of cliff that runs through this whole category, and it is worth naming because it keeps catching buyers out. Dropbox Dash switches plans at 100 users, where 100 seats cost $18,000 a year and 101 seats cost $42,420. Gemini Enterprise caps self-serve and resold accounts at 25 seats and invoiced accounts at 1,000. Microsoft caps its discounted Copilot Business offer at 300 users. The ceiling, not the sticker price, is what actually decides the bill for a growing company.

The practical consequence for ChatGPT Business is a step change in how you buy, not a step change in price per seat. At 200 seats you are paying a published $20 and you can add a seat in the admin console. At 201 you are in a sales cycle, a security review, an annual invoice and a negotiated credit pool. If you expect to cross 200 within the term you are about to sign, it is worth having that conversation before you commit, because the Business seats you bought are not refundable and annual additions stay committed to the end of the term.

We wrote about where the thresholds sit across the whole category in enterprise search software for teams over 100 users, and the ChatGPT specific migration math is in the ChatGPT Business 200 seat cap.

OpenAI publishes the credit rate card in full but does not publish what a credit costs in dollars. The rate card tells you how many credits a message, a deep research task or a million tokens consumes. It does not tell you the price of the credit itself, which is set in the Order Form for Enterprise and bought through the billing screen for Business. That is the single biggest gap in ChatGPT Enterprise pricing.

Credits pay for advanced models and features once the included allowance is gone. Core work does not touch them. In OpenAI's words: "Credits only apply to advanced models and features. You do not need credits to use core ChatGPT models and features such as search, file upload, and canvas." The published rates, read on September 2, 2026:

ChatGPT Work and Codex are priced on tokens instead, as credits per million: GPT-5.6 Sol is 100 credits in, 10 cached, 500 out; GPT-5.6 Terra is 50, 5 and 300; GPT-5.6 Luna is 5, 0.5 and 30. OpenAI notes that "A typical Codex task using GPT-5.6 Sol may consume between 5 and 30 credits per task" and that Sol's promotional rate is "available at least through November 21, 2026", which is a discount with a published expiry attached to it.

Three rules around credits are worth writing into your own model. Credits are not refundable except where the law requires it or an account compromise is confirmed. On Business they are valid for 12 months after purchase; on Enterprise and Edu, expiry is whatever the Order Form says. And when the pool empties on Enterprise, "advanced features will be paused unless Workspace Owners enable overages or purchases additional credits". One genuinely fair detail: if ChatGPT auto-routes a message to a mini model, that message costs zero credits.

Compare that with how the rest of the category meters. Amazon Q Business bills index capacity by the hour whether or not anyone searches, which puts a floor of roughly $193 a month under an enterprise index before a single query. Gemini Enterprise charges $5 per GiB per month over the edition allowance. Metered variable cost on top of a seat is now the norm, and the seat price is rarely the bill.

Business is self-serve, published at $20 or $100 per seat, capped at 200 seats, and gives every user a per seat usage allowance. Enterprise is sold by quote, invoiced annually, has no published seat ceiling, and replaces per seat allowances with one shared credit pool that the whole workspace draws from. The difference that matters day to day is how usage is rationed, not which features exist.

On Business, each person gets their own included limits and hits their own wall. If the workspace has bought credits, they keep going from the shared pool; if not, the feature blocks and they ask an owner for more. On Enterprise and Edu, there are no per seat caps by default at all. Everyone pulls from the contract level pool until it is empty, at which point advanced features pause across the workspace unless an owner has enabled overages. Owners can set a hard overage limit, and OpenAI notes that setting it to zero means the organization cannot go into overage at all.

There is one more Enterprise-only wrinkle that changes headcount math. Enterprise workspaces have two seat types: standard ChatGPT seats and Codex seats. Codex seats are Codex only, do not include ChatGPT access, and are described as "usage-based seats, with no fixed cost per user per month" that require workspace credits from first use. If a user on a Codex seat opens ChatGPT they are told it is not available for their seat type. Users cannot change their own seat type: "No. Seat types are assigned by workspace owners or admins."

A subtlety worth checking in your own contract: OpenAI states that "Enterprise and Edu plans without flexible pricing have the same per-seat usage limits as Plus for most features." An Enterprise contract does not automatically buy more usage than a $20 consumer plan. Flexible pricing is what buys that. If you are signing Enterprise mainly for capacity rather than for controls, confirm which model your Order Form puts you on.

Both plans get the governance features that no individual plan has. On the published feature matrix, SAML SSO, SCIM, Enterprise Key Management, the Compliance API logs platform, SOC 2 Type 2 compliance, the connector registry and Company Knowledge are all marked unavailable on Free, Go, Plus and Pro. If those are what you need, Business is the cheapest door in at $20 a seat, and it holds until seat 200.

ChatGPT Business Standard seats are limited per five hour window, and the allowance depends on the model: roughly 10 to 100 messages on GPT-5.6 Sol, 25 to 200 on Terra and 250 to 2,000 on Luna. Premium seats get five times that allowance and, importantly, no five hour window at all. Past the allowance, eligible usage continues only if the workspace has credits available and its spending controls allow it.

The ranges are wide on purpose and OpenAI explains why: "usage depends on the model, task size, reasoning effort, and amount of work performed. A longer task can use more of your allowance than a short task, so a fixed number of messages is not a reliable measure of remaining usage." That is honest, and it is also the reason a seat allowance is hard to budget against. Cloud chats can consume more of an allowance than local messages do.

Two operational notes for admins. A scheduled downgrade from Premium to Standard does not reduce access before its effective date, so you can plan seat changes ahead of a renewal without cutting someone off early. And hitting a limit in one place does not lock everything: "Reaching a limit for one experience does not mean every ChatGPT feature is unavailable."

Image generation deserves a separate line because it burns allowance faster than it looks. OpenAI says image generations use included limits roughly three to five times faster than comparable turns without them, depending on quality and size, and then draw on credits after that. Speed settings do the same thing: fast mode consumes credits at a higher rate on supported models.

At the same $20 a month, ChatGPT Business buys governance that Plus does not have at any price. SAML SSO, SCIM provisioning, the connector registry, Company Knowledge, the Compliance API logs platform, Enterprise Key Management and SOC 2 Type 2 compliance are all listed as unavailable on Free, Go, Plus and Pro. If your reason for buying is administrative control rather than raw usage, Business is the same money and a different product.

If your reason for buying is capacity, the comparison is closer than the tiers suggest. A Business Standard seat and a Plus seat are both $20 on annual billing, and both are governed by five hour windows. What Business adds on capacity is the option to buy credits into a shared pool so that one person's heavy week does not simply stop them, plus the Premium seat at $100 for five times the usage with no window.

Where it stops being a close call is headcount and compliance. Plus is licensed for individuals; OpenAI says so directly on the pricing page. Running a company on individually expensed Plus accounts means no SSO, no provisioning or deprovisioning when someone leaves, no workspace-level audit, and no central control over which connected apps are enabled. That is a security review you will eventually fail, and it usually surfaces during the first customer questionnaire rather than internally.

The honest counter-argument is that Business is not enterprise search and does not claim to be. Connected apps in ChatGPT are authorized by each individual user, one account at a time, and the assistant inherits that person's permissions from the source. That is a sensible security model and it is also why coverage is per person and opt-in rather than organization-wide. If what you need is one index across every app that every employee can query, the next section is the relevant one.

Not on its own, and OpenAI does not really pitch it as a replacement. ChatGPT connects to services through apps and plugins that each user authorizes individually, so what the assistant can reach is the union of what that one person has connected and is allowed to open. That is a good security posture and a poor coverage model. Enterprise search inverts it: one governed index over the company's content, with each person's permissions applied at query time.

OpenAI documents the mechanism clearly. Apps connect ChatGPT to services like Google Drive and Slack so you can find a file, summarize a document or look up a message. As of July 9, 2026 the app directory migrated into a plugin directory, with workspace admins managing installation in Workspace settings and configuring each underlying app's access and permissions. Some apps support syncing to index content in advance for faster responses. Permissions default to "Important actions", which allows reading automatically but asks before anything with meaningful effect outside ChatGPT.

Read that carefully and the shape of the product is obvious. It is an assistant with delegated access, not a company index. Admins decide which apps may be installed; individual users still connect their own accounts. Nobody in the organization can ask a question whose answer sits in a document they have not personally connected, and nothing gives an admin a single view of what is searchable company-wide.

This is the same conclusion we reached about Claude Cowork, and about Gemini Enterprise, Glean and Copilot. The general-purpose assistants are converging on the same design because it is the right one for an assistant. It just does not answer the question a new hire asks in week two, which is what our renewal terms are, where the security questionnaire template lives, and which of the four conflicting onboarding docs is current.

That is the job unified search does. InSearch indexes Drive, Slack, Notion, Confluence, Gmail, Jira and Salesforce together, returns a written answer with inline citations to the source documents, and enforces each source's native access controls at query time so a company-wide rollout never surfaces something a person could not already open. It is $20 per user per month on Team billed annually, with no 200 seat ceiling. Plenty of teams run both: ChatGPT for drafting and reasoning, a search layer for finding what the company already knows. They are not the same purchase.