Claude Enterprise pricing, Claude Team vs Enterprise, and Claude Enterprise cost per user

Twenty dollars per user per month, billed annually. That is the published figure on claude.com/pricing, where the Enterprise card reads "Seat price + usage at API rates" and then "$20/seat. Usage cost scales with model and task." The seat is annual only. There is no monthly Enterprise billing cycle, and there is no free trial unless you buy through Anthropic sales.

The sentence that decides your budget is in the Enterprise billing article rather than on the pricing page: "The seat fee gives each person access to Claude on web, desktop, and mobile, plus Claude Code and Cowork, but doesn't include any usage. Every token your team consumes is billed separately at standard API rates." A Claude Enterprise seat is a license to be metered.

That is not a criticism, and for some organizations it is genuinely the cheaper model. A team of researchers who each run two long sessions a week will spend less at API rates than they would on a bundled plan sized for heavy users. The problem is planning. A seat based plan gives you a number you can multiply by headcount and put in a budget. A usage based plan gives you a number that depends on which model your people pick from a dropdown, and the spread between those models is five to one on input tokens and five to one on output tokens.

Here is the arithmetic, using Anthropic's own published API rates and one stated assumption. Suppose a single person consumes 2 million input tokens and 400,000 output tokens in a month. That is a working professional using Claude daily, not an engineer running agents overnight.

The token volumes are our assumption and we have labelled them as such. The rates are Anthropic's, published on the same page as the seat price. The point is the shape rather than the total: on the same workload, the model choice swings the all in cost per user from $24 to $60, and the seat fee is the smallest term in every row. If your organization turns on Fable and does not set spend limits, the seat price stops being a useful planning number.

Two multipliers sit on top of that. Prompt caching is billed separately, at $0.25 per MTok to read and $12.50 per MTok to write on Fable 5.1. And if your organization enables the US only inference setting for data residency, Anthropic bills usage at 1.1x standard API rates. Regulated buyers who need US only processing should add ten percent to every usage figure they model.

Yes. Twenty seats. Anthropic states it verbatim in the Team to Enterprise migration guide, in the step where you choose your seat count: "Add the number of seats needed for all your team members (Enterprise organizations have a 20 seat minimum)." At $20 a seat billed annually that puts the floor for Claude Enterprise at $4,800 a year in seat fees, before any usage is billed at all.

This is the single most useful number on this page and it is missing from almost every Claude Enterprise pricing article in circulation, because it is not on the pricing page. It is buried in a migration how to. If you have 12 people and you want audit logs, SCIM or the Compliance API, you are not buying 12 Enterprise seats. You are buying 20.

The second commitment detail sits in the billing article: "Seats cannot be removed mid-term on self-serve Enterprise plans." You are charged for the seat count at the start of the annual cycle, and adding seats during the term triggers an immediate prorated charge for the remainder of the year. Anthropic works the example through: an annual plan starting January 1 with 50 seats is charged up front, and 10 seats added on April 1 are charged immediately for the remaining months.

Put the two together and the real Enterprise floor is a 12 month, 20 seat, annual commitment with no downward adjustment inside the term. That is a normal enterprise software shape. It is just a long way from the impression created by a $20 sticker sitting next to a $17 individual plan.

One more thing worth checking before you commit: migrating a Team organization to self serve Enterprise is one way. Anthropic marks it clearly, in bold, on its own page: "Migrating an organization from Team to Enterprise via this pathway is not reversible." If you are unsure, the sales assisted route is the one that carries trials.

Nothing, on price. A Claude Team standard seat and a Claude Enterprise seat are both $20 per user per month billed annually. What changes is that Team includes usage inside published limits and Enterprise includes none, and that Enterprise unlocks eleven governance features Team does not have. Claude Team also caps at 150 people, so above that the choice is made for you.

This is the delta, taken row by row from Anthropic's own three way comparison table on September 6, 2026. Only the rows that actually differ are listed, because the twenty odd rows where Team and Enterprise are identical are not a reason to upgrade.

Read that list as two separate purchases. Rows one and two are capability: a bigger context window and the top model. Rows three through eleven are governance: the things a security review asks for. If nobody in your organization is asking for audit logs, SCIM or a retention policy, and 200k of context is enough, Team at the same seat price with usage included is the better commercial deal and the more predictable one.

If you do need the governance rows, price the whole thing rather than the seat. Twenty seats minimum, annual only, no mid term removals, and a usage meter with no included allowance.

They are the same product at the same seat price, bought two different ways, and the difference shows up in how usage is funded and what happens when the money runs out. Self serve buys usage up front in credits and stops the whole organization at zero. Sales assisted meters usage and invoices it monthly in arrears, so nobody is ever blocked, but there is no ceiling until you set one.

Anthropic describes the self serve failure mode plainly: "Everyone's usage draws from the same credit balance. One person's heavy usage depletes credits faster for everyone, so monitoring and spend limits matter here. When the balance hits zero, usage stops for the whole organization until an Owner purchases more." One person running a long agent job on a Friday afternoon can take the tool away from everybody until somebody with the billing permission logs in.

Sales assisted has the opposite property: "There's no balance to deplete... One person's heavy usage doesn't block anyone else, it just increases what shows up on the bill." Which of those two you prefer is a real decision, and it is closer to a treasury question than a software one.

Spend limits exist on both, at organization and individual level, and Anthropic notes they work hierarchically so a user cannot exceed whichever limit is lower. There is one caveat worth quoting because it surprises people: "Owners can set limits to 'unlimited,' but consumption will still be billed. You cannot disable billing for usage, usage-based pricing is a core part of the Enterprise plan."

The rest of the split is procurement rather than product. Anthropic's own comparison table marks four things as sales assisted only: AWS Marketplace availability, consultation, tiered incentives on committed spend, and trials. There is no trial on self serve Enterprise. Invoicing and net terms are also sales assisted only, and self serve Enterprise is billed in USD only, so a company that needs to pay in another currency has to go through sales regardless of size.

Five capabilities that are on by default for Team are off by default on Enterprise: Skills (including skill creation and sharing), code execution and file creation, interactive content in artifacts, Claude Design, and Claude in Chrome. Nothing is deleted, but on the morning after a migration your users will find features missing until an admin turns them back on.

Anthropic lists all five in the migration guide and is explicit that the underlying content survives: "if your Team plan had skills enabled and members had created custom skills, those skills are not deleted during the migration, they're retained in your organization. When an admin re-enables skills on the Enterprise plan, any previously created custom skills will be available again."

Three other migration details are worth putting on a checklist, because each one has generated a support ticket for somebody:

• Seat assignment gaps. Anthropic warns that after migration some users may show as "No seat assigned" rather than being mapped automatically, and a user without a seat has no access until an admin fixes it. Check your heaviest users first.
• Spend limit lockouts before you even move. If you set an organization level spend limit on Team, raise it before go live: "Once the org-level spend limit is reached, all users lose access immediately until the spend limit is raised or the new month begins."
• Cutover downtime. On the sales assisted path, users cannot access Claude while provisioning runs, and Anthropic recommends planning the rollout for the next business morning rather than mid workday. Everyone should log out and back in afterwards to clear cached legacy settings.

There is one more thing that no competing article seems to have noticed yet, and it invalidates a lot of published comparison content. The Standard and Premium Enterprise seat types are now legacy. So are the older Chat and Chat + Claude Code seats. Anthropic states that both "are no longer available for new contracts" and that they transition to a single Enterprise seat at the next renewal. If you are reading an article that describes choosing between Standard and Premium seats on Claude Enterprise, it is describing a plan you can no longer buy. Standard and Premium still exist on Claude Team, which is where the confusion comes from.

Claude Enterprise is an assistant sold per person and metered per token. Enterprise search is an index sold per person at a flat rate. They overlap enough to appear on the same shortlist and they solve different halves of the problem, so the honest test is whether your people need to think with a model or find something somebody already wrote.

Give Anthropic its due here. Claude lists enterprise search as available on Team and Enterprise, connectors are real, and the permission model is handled correctly: an admin enables a connector for the organization, then each person authorizes their own account, and Claude inherits that person's permissions from the source. Nobody sees anything they could not already open. That is the right design and several better known products get it wrong.

The consequence of that design is what matters commercially. Because each connector is authorized per person, coverage is per person and opt in. In a 400 person company with five connected systems, complete coverage means 2,000 individual OAuth grants that no administrator can complete on anyone's behalf. In practice you get an assistant that knows a great deal about whatever each individual happened to connect, and nothing about the rest of the company. That is a fine outcome if what you wanted was a personal assistant. It is not an organizational index, and it will not answer "what did we tell this customer about renewal terms in March" for someone who was not in the thread.

The second consequence is the bill. An assistant that thinks harder costs more, by design, and the useful work is exactly the expensive work. A search index does not have that property: retrieval cost per query barely moves whether the answer is easy or hard, which is why search can be sold flat.

InSearch does the second job. One permission aware index across Google Drive, Slack, Notion, Confluence, Gmail, Jira and Salesforce, one question in plain English, and one written answer with inline citations back to the source document. Connected once by an administrator rather than 2,000 times by individuals, priced per seat with no token meter, and never trained on your data. Plenty of companies run both, and that is a reasonable answer too. Just do not buy the assistant expecting the index, or budget for the index and get billed for the assistant.

The console at the top of this page runs against real content, so you can see what a cited cross app answer looks like before anyone talks to sales.