Claude Enterprise is Anthropic's per-seat, usage-based plan for organisations: a 20-seat minimum on self-serve, a single seat type billed annually, and every token your team spends billed at standard API rates on top of the seat fee. The hard part of the rollout is not picking the plan. It is that two of the earliest steps — migrating an existing Team workspace, and claiming your company domain — cannot be undone. Decide those before you buy seats.
We are an AI implementation firm, so read this with that interest in mind. It is the page we could not find: page one explains what the plan costs and how it compares to Pro or Team, and none of it explains the order of operations.
What Claude Enterprise actually includes
Verified on 6 August 2026 against Anthropic's own help centre and pricing page. Anthropic describes Enterprise as having "a single seat type, priced per user per month and billed annually," with a minimum of 20 seats on the self-serve path and 50 through sales.
The seat fee is not the whole bill. Anthropic states that "every token your team uses… is billed at standard API rates on top of your seat cost," and that usage-based Enterprise plans have "no plan or seat-level usage limits." That is the most consequential fact here, and we return to it below.
Included in the seat fee: Claude Code and Cowork, plus connectors for Google Drive, Gmail, Google Calendar, GitHub, Microsoft 365 and Slack. The admin surface adds SCIM provisioning, audit logs that "capture key information about user actions, system events, and data access," custom data retention controls, a Compliance API, an Analytics API, customer-managed encryption keys held in your own cloud provider, US-only inference, IP allowlisting, and a HIPAA-ready configuration under a BAA.
For contrast, published Team pricing is $20 per seat per month billed annually ($25 monthly) for a standard seat and $100 ($125 monthly) for a premium seat, sized for teams of 2–150, and it already includes SSO, admin controls and central billing. Since our first check, Anthropic has added an Enterprise seat price to its pricing page: $20 per seat per month, billed annually, with a 20-seat minimum, plus usage at API rates (rechecked 20 October 2026).
What is the Claude Enterprise minimum seat count?
The Claude Enterprise minimum is 20 seats on a self-serve plan and 50 seats on a sales-assisted plan, according to Anthropic's help centre on 6 November 2026. Anthropic says both paths have the same features and the same seat pricing. At the listed $20 per seat per month, billed annually, that puts the seat floor at $4,800 a year for 20 seats and $12,000 a year for 50, before any usage, which is billed at API rates on top.
| Self-serve Enterprise | Sales-assisted Enterprise | |
|---|---|---|
| Minimum seats | 20 | 50 |
| Seat floor at $20/seat/month, annual | $4,800 a year | $12,000 a year |
| Payment | Credit card, ACH | Credit card, ACH, invoicing |
| Currency | USD only | Multi-currency |
| Usage billing | Credits purchased upfront | Billed monthly in arrears |
| Reducing seats | Only at renewal | Through your account manager |
If fewer than 20 people need Claude, the published option is Team, sold for 2 to 150 seats. If you need Enterprise controls such as SCIM or audit logs for a smaller group, you still pay for 20 seats.
On self-serve, Anthropic's seat article says you can add seats at any time but "seats cannot be removed during your annual term." We recommend buying the minimum, or the pilot group if it is larger, and adding seats team by team. Buy them before you switch on SCIM, because provisioning stalls when no seat is free (see what a Claude seat is and how provisioning uses it).
Claude Team vs Enterprise: which plan fits?
Choose Claude Team if you have 150 people or fewer and SSO is enough identity control; choose Enterprise once you need SCIM deprovisioning, audit logs, custom data retention or a HIPAA-ready setup. Anthropic now lists the same $20 annual price for a Team standard seat and an Enterprise seat, so the cost difference comes from usage. A Team seat includes usage up to its limits. An Enterprise seat has no seat-level limit, and every token is billed at API rates on top.
| Claude Team | Claude Enterprise | |
|---|---|---|
| Seat price | Standard $20 per seat per month billed annually ($25 monthly); Premium $100 ($125 monthly) | $20 per seat per month, billed annually, plus usage at API rates |
| Size | 2 to 150 people | 20-seat minimum on self-serve; no maximum listed |
| Usage | Included up to each seat's limits; a Premium seat gets 5x a Standard seat | No seat-level limits; all usage billed at API rates |
| Pooled API credits | Up to $500 per month | None |
| Billing | Monthly or annual, credit card | Annual only; card or ACH on self-serve, invoicing through sales |
| SSO and domain verification | Yes | Yes |
| SCIM provisioning | No | Yes |
| Role-based access | No | Yes |
| Audit logs and Compliance API | No | Yes |
| Custom data retention | No | Yes |
| Domain capture | No | Yes |
| HIPAA-ready offering | No | Yes |
| Claude Code and Cowork | Included | Included |
Source: Anthropic's pricing page and plan comparison table, checked 20 October 2026.
The rule we apply: pick the plan by the controls your security review requires, then check the bill. If IT must be able to deprovision people from the identity provider, or legal needs audit logs and a retention setting, Team cannot provide them at any seat count, and the comparison ends there. If neither applies and you are under 150 people, start on Team. You can move to Enterprise later, but the move cannot be undone (next section).
On cost, the two plans handle heavy users differently. Team caps usage per seat, so heavy Claude Code users hit their limits and need Premium seats at five times the Standard price. Enterprise never caps a seat, so the same heavy users appear as a larger token bill. For staff who mostly chat, Team gives the more predictable bill. For a team that will run Claude Code most of the day, price both plans from a few weeks of measured usage before you sign, because the seat price alone will not tell you which is cheaper.
Both plans include Claude Code and Cowork. If you are still deciding which of the two your staff should use, see our comparison of Claude Cowork and Claude Code.
Claude Duet, the proactive agent Anthropic released in research preview on 21 October 2026, is not part of Team or Enterprise yet: Anthropic says those plans will come later. Our guide to Claude Duet vs Cowork covers what it does and who can use it today.
The two doors that do not open again
Plan selection is a spreadsheet exercise, and reversible in the direction most buyers worry about: you can start on Team and move up. The decisions that determine whether the rollout works sit downstream of it, and two are permanent — flagged, to Anthropic's credit, but inside help-centre articles nobody reads until they are halfway through the change.
Migrating a Team workspace to Enterprise
If you already run Claude on Team, the upgrade path is a migration, not a plan toggle. Anthropic states plainly: "Migrating an organization from Team to Enterprise via this pathway is not reversible."
What carries over is generous: "all chat history and conversations," "projects and shared content," and "user memberships and roles," with unused Team credits rolling into the new plan. What surprises people is the feature defaults. Several capabilities that are on by default under Team arrive disabled under Enterprise: Skills (and by dependency skill creation and sharing), code execution and file creation, interactive content in artifacts, Claude Design, and Claude in Chrome. The data is preserved and an admin can re-enable each one, but migrate on a Friday without telling anyone and a meaningful number of people will open Claude on Monday to find that the thing they built their week around has gone quiet.
Decide in advance which of the five you will re-enable, and turn them back on in the same maintenance window rather than one ticket at a time.
Domain capture
The second door is sharper. Domain capture lets an Enterprise admin discover and pull existing personal Claude accounts on a verified company domain into the corporate workspace. Anthropic's warning is one sentence: "Domain capture is a one-way door. Once enabled, it can't be reversed."
The prerequisites are real work: DNS domain verification, SSO enforced rather than merely configured, JIT provisioning or SCIM enabled, and organisation creation restricted on the verified domain. Once you initiate the claim the window is 30 days, and Anthropic is explicit: "the migration deadline is a single org-wide date, not a rolling per-user timer. All affected accounts share the same deadline."
Inside that window each person chooses to bring existing chats, projects, files and memory into a new account in your workspace, or to start clean. After it closes, unmigrated personal accounts are deactivated, and paid Pro and Max subscriptions are cancelled with a prorated refund.
That last clause is where the people problem lives. Some of your strongest users have paid for Claude themselves for months, with a year of working context in it. A domain claim announced by email on day one of a 30-day clock, with no mention of the merge option, reliably produces a small group of angry senior individual contributors, the exact people whose adoption you needed. The mechanism is fine; the communication is what fails.
A rollout sequence that survives contact with IT
Here is the order we use, with the irreversible steps marked. Nothing in it is exotic; the value is in doing the permanent things deliberately.
Both one-way doors sit early, so the pressure to move fast in week one points straight at the two decisions that most deserve a week of thought. And roles and spend limits come before broad invitations. Retrofitting permissions onto a live workspace is far worse than setting them while it is small.
Identity comes before everything else
SSO enforcement, SCIM or JIT provisioning and DNS domain verification are prerequisites for domain capture, so they get done regardless. Doing them first also gives you a clean answer to "who has an account", the question your security review opens with, and the one most organisations cannot answer.
SCIM matters more than it looks. Deprovisioning through your identity provider is the difference between an offboarding checklist that works and one that quietly leaves former employees holding a session against your workspace. If you need to argue Enterprise over Team internally, that beats the feature list.
What is a seat in Claude, and what does "No seat assigned" mean?
A seat is the paid licence that lets one member of a Team or Enterprise organisation use Claude. A member set to "No seat assigned" stays in the organisation and keeps their account, but cannot use Claude until an admin assigns a seat again. Anthropic's help centre articles on Team seats and Enterprise seats (checked 20 October 2026) add four rules that matter during a rollout:
- Only Owners and Primary Owners can buy seats. Admins and above can move members between seat types, including to "No seat assigned".
- Members with no seat do not count toward the seat limit. Pending invitations do.
- When JIT or SCIM provisioning finds no free seat, provisioning does not complete until an Owner buys another one.
- Removing a member frees their seat for someone else and leaves the bill unchanged. On a self-serve Enterprise plan you cannot remove seats during the annual term; reductions take effect at renewal.
The provisioning rule matters most on launch day. Switch on SCIM before you have bought enough seats and new starters sync from your identity provider, but their provisioning does not complete until someone buys more seats. Count the people your identity group will push, buy the seats, then enable provisioning.
Set roles and spend limits before you invite people
Enterprise custom roles control which capabilities a member can reach. Anthropic describes each as "a set of permissions that grant or restrict access to specific capabilities like chat, Claude Cowork, Claude Code, and web search, plus the connectors your organization has added." Seven admin areas can also be delegated without making anyone an Owner: Identity & Access, Billing, Analytics, Privacy, User Management, Libraries and Directory management, each at "can view" or "can manage."
Two behaviours will catch you out if you design permissions the way you design them elsewhere.
Permissions are additive across groups. If a member belongs to multiple groups with different custom roles, Anthropic's documentation says "they get the union of all permissions from all roles in their chain." You cannot use a restrictive role to claw back access another grants. If your model is "assign the locked-down role to contractors," it stops behaving as expected the moment a contractor is also in the marketing group.
Spend limits invert that logic. Group limits assign a per-user monthly ceiling across a group, but "individual limits always override group limits, regardless of which is higher", so an individual exception cuts both ways. The org-wide limit is a hard ceiling, and Anthropic warns that hitting it "can be disruptive" because it affects everyone simultaneously. When a limit bites, Claude "will complete the current turn before limiting further usage," so nobody loses work mid-task.
Budget for propagation too: role changes can take up to 15 minutes to apply, long enough that an admin testing one concludes it failed and changes something else.
What you can actually configure on data
The data controls are why most regulated buyers move from Team to Enterprise, and they are worth naming precisely: custom data retention controls, audit logs, the Compliance API for programmatic access to usage data and activity logs, customer-managed encryption keys in your own cloud provider, US-only inference, IP allowlisting, tenant restrictions for network-level access control, and a HIPAA-ready configuration with a BAA.
Bring security in at the design stage, not sign-off. On our Claude for business engagements the governance and security review runs as one of three parallel workstreams — value, governance, enablement — so the control conversation is not a veto arriving after the pilot is staffed. The controls that review covers are mapped to NIST AI RMF and the OWASP LLM Top 10 in our AI implementation risk checklist.
Your bill is a function of behaviour, not headcount
Because usage is billed at API rates on top of seats, seat count tells you the floor and nothing else. Anthropic is direct: "usage is measured in tokens. Long prompts and long conversations consume more tokens," and consumption varies sharply by surface: "a single Cowork task or Claude Code debug session can consume many more tokens than chat." The pool is shared across your organisation.
This is the cost structure we describe on our AI agent cost guide: spend splits into usage, runtime and operations, and the useful unit is cost per workflow rather than cost per person. The drivers are session design, context size, tool-call frequency and model routing. Anthropic's own levers line up: set an org default to a lower-consumption model, restrict model access by role, cap effort levels, watch the analytics. For the per-task arithmetic behind those levers, see our Claude API pricing guide for business.
So do not forecast this plan from a headcount spreadsheet. Forecast from a few instrumented weeks, then review weekly. Monthly is too slow to catch context bloat before it compounds.
We publish no rate card for implementation work and will not invent a range for a rollout we have not scoped. The shape we can give: a first deployment typically runs 6–12 weeks — strategy and readiness, applied build with a human-in-the-loop QA gate, then production delivery with ownership, observability and exception handling.
Choose the pilot by the measurement, not the enthusiasm
The instinct is to pilot with whoever asked loudest. The better filter: which team already has a number you could move, measured the same way twice, before Claude arrives?
Engineering is the usual first candidate, because Claude Code is in the seat fee and the work is legible: cycle time, review turnaround, time to first passing build. Our Claude Code for companies explorer maps how the same rollout looks across departments, from engineering and operations through to QA and analytics, which helps you see where the second and third waves land. Once engineering is picked, our Claude Code adoption plan for engineering teams covers what to standardise in the repository before usage spreads: the shared instructions file, the permission rules and the check that proves an automated change landed.
Whatever you pick, run one pilot before wider rollout and write the metric down before the first seat is provisioned. A baseline captured after people start using the tool is not one.
Where these rollouts stall
Four failure modes, in rough order of frequency.
The door gets opened by accident. An admin enabling domain capture to "tidy up accounts" starts a 30-day org-wide clock nobody planned for. Make both irreversible steps require two named approvers.
Nobody owns the bill. Usage is uncapped at the plan level and the org-wide ceiling is blunt. Without group limits and a weekly review, the first surprising invoice arrives before anyone has a per-workflow number to explain it.
Permissions were designed subtractively. Teams model access as "everyone gets the base role, contractors get the restricted one," then meet the union rule. Design additively: build the narrow role and add capabilities.
The pilot has no denominator. Six weeks in, the team reports that Claude is "really helpful" and finance asks what changed. If you cannot answer in the same unit you used before the rollout, the second budget conversation is much harder than the first.
None of these are product problems. All four are sequencing and ownership problems — which is why we treat this as a governance project that happens to involve a model.
How do I find and verify an Anthropic implementation partner?
Search Anthropic's Claude Partner Network directory at partnerhub.claude.com/directory, filter by Partner Tier (Select, Preferred or Global Premier), and check the profile's tier matches the firm's proposal. Then ask which people on the proposed team hold a current Anthropic certification, and check each one. The profile's description is written by the partner, so a tier named only in that text proves nothing.
When we checked profiles in October 2026 they showed the tier but not the firm's certified headcount or number of customer deployments. Ask for both, plus one named reference you can call. Our guide to Anthropic's partner program tiers covers what each tier requires and how firms gain and lose one.
Frequently asked questions
Can we buy fewer than 20 Claude Enterprise seats?
No. Self-serve Enterprise starts at 20 seats and sales-assisted Enterprise at 50, per Anthropic's help centre, rechecked 6 November 2026. A smaller group can use Team, which Anthropic sells for 2 to 150 seats.
How much does Claude Enterprise cost?
$20 per seat per month, billed annually, with a 20-seat minimum on self-serve, plus usage billed at standard API rates, according to Anthropic's pricing page on 20 October 2026. Usage-based Enterprise plans have no plan or seat-level usage limits, so the usage line is the part to model. Buyers who go through sales can ask for a tailored quote.
Can we move from Team to Enterprise later?
Yes. The migration keeps chat history, projects and shared content, memberships and roles, and rolls over unused credits. But Anthropic states it is not reversible, and several capabilities enabled by default on Team — Skills, code execution and file creation, interactive artifacts, Claude Design and Claude in Chrome — arrive disabled on Enterprise until an admin turns them back on.
What does "not an Enterprise or a Team organization" mean?
Anthropic's help centre does not document this exact message (we searched on 20 October 2026). Its wording says the feature you tried is limited to Team and Enterprise organisations and the account you are using is not in one. Check first that you are signed in with your work account and not a personal Free, Pro or Max one. In Claude Code, Anthropic's documented fix for the wrong account is to run /logout, sign in again and choose the work account.
Who needs to complete the setup before team members can use enterprise search?
An Owner of the organisation. Anthropic's help centre says Owners must complete the initial setup before other members can use enterprise search, which is available on both Team and Enterprise plans (checked 20 October 2026).
Do we have to enable domain capture?
No. It is opt-in, for organisations that want to guarantee no non-Enterprise accounts exist on their verified domain. Enabling it cannot be reversed, and it starts a single 30-day org-wide window after which unmigrated personal accounts are deactivated and any Pro or Max subscriptions are cancelled with a prorated refund.
Is Claude Code included in an Enterprise seat?
Yes — Anthropic lists Claude Code and Cowork as included in the seat fee. Both consume substantially more tokens than chat, so seat inclusion does not make them free in the usage line of your bill.
How long should a first Claude Enterprise rollout take?
Do identity and permissions work before any broad invitation, then one measured pilot. In our implementation work a first deployment typically runs 6–12 weeks. If someone quotes you a two-week enterprise rollout, ask which irreversible step they intend to skip.
A note on who is telling you this
ATI holds and claims no tier in Anthropic's Claude Partner Network, and we are not asking you to buy on a badge. We have written separately on what the partner tiers do and do not tell you and how to evaluate a Claude implementation partner. Every product fact above comes from Anthropic's own published pages. Most are dated 6 August 2026; the plan comparison, Enterprise price, seat rules and partner directory details were rechecked on 20 October 2026. Plan mechanics change, so verify anything load-bearing before you sign.
If you want a second pair of eyes on the sequence before you open either door, book a call and we will walk it against your identity setup and first candidate workflow.