Claude Enterprise Pricing, Explained

What does Claude Enterprise cost?

Claude Enterprise pricing has two parts, and only one of them is published. As of October 2026 the seat is listed at $20 per user per month, billed annually. Usage is billed separately: every token your team spends across Claude's surfaces, Chat and Claude Code included, is charged at standard API (application programming interface) rates on top of the seat, with no allowance bundled in. Your seat line is fixed by purchased seat count. Your usage line is fixed by nothing.

That is the shape of the bill. What follows is who qualifies, what the seat buys over the Team plan, and how the number you are quoted gets built.

Why Claude Enterprise pricing has a seat price but no list price

An Enterprise seat buys permission to use Claude under your company's controls. It carries no tokens with it. No published monthly total exists because two seats can differ by a hundred times: a manager who asks four questions a day and an engineer running Claude Code across a large repository both cost one seat, and their token bills are nothing alike. That opacity is not unique to Anthropic, though it is unevenly distributed: only one of the two vendors publishes an enterprise seat price at all, which is worth knowing before you try to lay two quotes side by side.

How you pay depends on which door you walk through. The self-serve path takes a card or ACH bank transfer and runs on credits bought upfront, so usage draws down a balance you top up. The sales-assisted path bills usage monthly in arrears. Self-serve pays by card or ACH only, at any size. Sales-assisted invoices can be paid by bank transfer, or by card when they are smaller. The $50,000 figure you may have seen is a payment-method rule, not an eligibility floor: as of October 2026, invoices of $50,000 or more can only be paid by bank transfer (ACH or wire). Enterprise is also sold through the AWS Marketplace, billed monthly in arrears like the sales-assisted path. If you already have an AWS agreement, ask your account team whether Marketplace spend counts against your committed spend, because that answer lives in your AWS contract rather than in Anthropic's price list.

How many seats does Claude Enterprise require?

There are two minimums. Self-serve Enterprise starts at 20 seats. Sales-assisted Enterprise starts at 50 seats. These are purchased-seat minimums, not headcount requirements: a smaller team can buy 20 seats and leave some unused. Removing a member leaves the purchased seat available and does not reduce the bill. Team runs from 2 to 150 seats, so an organisation that needs more than 150 seats moves to Enterprise because Team has run out of room, not because of any single control. The cap counts the seats you buy, not the people you employ: a company of 2,000 that puts 90 of them on Claude can stay on Team.

[
  {
    "plan": "Pro",
    "usd_per_seat_month": 17,
    "min_seats": 1
  },
  {
    "plan": "Team standard",
    "usd_per_seat_month": 20,
    "min_seats": 2
  },
  {
    "plan": "Team premium",
    "usd_per_seat_month": 100,
    "min_seats": 2
  },
  {
    "plan": "Enterprise self-serve",
    "usd_per_seat_month": 20,
    "min_seats": 20
  },
  {
    "plan": "Enterprise sales-assisted",
    "usd_per_seat_month": 20,
    "min_seats": 50
  }
]

Read that chart with one thing in mind. A Team premium seat at $100 per month carries five times the usage of a standard Team seat, and that usage is included. An Enterprise seat at $20 per month includes none. The smaller number is the one with the open end.

What Enterprise buys that Team does not

Start with what Team already has, because this is where upgrade arguments usually go wrong. Team includes SSO (single sign-on), domain verification, JIT (just-in-time) provisioning, Owner, Admin and User roles, and spend controls at both the user and the organisation level. Single sign-on is not a reason to move to Enterprise. It is already there.

Enterprise adds the controls a security team asks for by name:

• SCIM (system for cross-domain identity management) provisioning, so a leaver is deprovisioned by your identity provider instead of by hand.
• Audit logs, which Anthropic lists as Enterprise only. OpenTelemetry monitoring for Claude Cowork is not on that list: as of October 2026 it is available on Team as well as Enterprise.
• A Compliance API that returns usage data, activity logs, chat histories and file content.
• Custom data retention controls.
• Domain capture and domain claiming, so existing accounts on your company's domain can be brought into the organisation. Team can verify a domain and block new personal accounts, but its admins cannot claim or migrate existing ones; members can migrate their own accounts voluntarily.
• Network-level access control and IP allowlisting.
• Customer-managed encryption keys. You provision a key in your own cloud account, and Anthropic uses that key to protect your team's chats, projects and files.
• US-only inference, which keeps your company's inference inside the United States.
• A HIPAA-ready offering (Health Insurance Portability and Accountability Act), plus fine grained role-based access control.

One idea runs through that list: evidence. Team lets a person use Claude under sane defaults. Enterprise lets you prove afterwards what that person did. When an auditor asks which files an employee uploaded last quarter, the Compliance API answers it and a Team plan cannot, because Team exposes no audit log and no endpoint to query one.

Data retention and model training

Content on commercial plans is not used to train Anthropic's models by default. That holds on Team as well as Enterprise, so "they will train on our conversations" is not a reason to upgrade. What Enterprise adds is control over how long content is kept, and the ability to read it back programmatically.

For its commercial products, Anthropic lists SOC 2 Type I and Type II reports, ISO 27001 and ISO/IEC 42001 certification, and its Data Processing Addendum, part of the commercial terms, covers GDPR (General Data Protection Regulation). Those are properties of the service and not line items you negotiate. Retention length, the HIPAA offering, customer-managed keys and the inference region are the parts that go into a contract.

The context window is not the reason anymore

Older write-ups describe an expanded 500K token context window as the Enterprise headline. Check that claim against the current help center before you buy on it. As of October 2026 Anthropic documents a 1M token context window for Opus 5.5, Opus 5, Sonnet 5.5 and Sonnet 5 when chatting with Claude on all paid plans, and 500K for Opus 4.8 and Sonnet 4.6. Claude Code supports 1M for those older models too, with usage credits required for Sonnet 4.6 except on usage-based Enterprise plans. Context length follows the model and surface you pick, so check both before buying for a bigger window. Choosing between them is its own question, covered in which Claude model to use for which job.

What decides your quote

Four inputs move the number.

• Seat count. Under 20 purchased seats Enterprise is not sold at all. From 20 to 49 you are on the self-serve path with fixed terms. At 50 and above you can still buy self-serve with no sales conversation, or go through a salesperson, and the sales-assisted path is where custom terms become possible.
• Term. Usage-based Enterprise uses a single seat type, priced per user per month and billed annually, except for HIPAA-ready organisations, which use separate Chat and Chat + Claude Code seats. The annual commitment is the default shape of the deal rather than a discount you have to ask for. That is a real change from the plans your people are on today, where cancelling takes effect at the end of the billing period you already paid for, so treat the year as a decision you make once rather than one you can walk back in April.
• Expected usage. Usage bills at API rates with no allowance, so this is where your risk sits. Estimate it from a real month of Team or API usage, not from headcount, because headcount tells you nothing about tokens. Which model your people reach for moves that figure as much as how much they type, so price the mix you expect: what Fable 5 costs per million tokens and the jobs it suits is the sort of per-model number that belongs in the estimate.
• The extras. HIPAA readiness, customer-managed encryption keys, US-only inference and a custom retention length are separate asks. Raise them on the first call, because they change which agreement you end up signing. US-only inference also changes the token price: as of October 2026 it bills usage at 1.1 times standard API rates for Opus 4.6, Sonnet 4.6 and later models.

Before that call, pull one real number: the tokens your people spent last month. Where Claude Code tokens actually go and how Claude usage limits work both explain how to read that figure. A quote you cannot check is a quote you cannot argue with.

Solo, small team, or org

• One person. Pro or Max. Enterprise requires buying at least 20 seats, even if you use only one. Start from choosing between the Claude plans. If you are not yet convinced a paid tier earns its keep, decide it from how often the free limit actually stops you mid-task. If Pro looks like the answer, check what Pro costs and where its usage limits stop you before you commit to a year of it. If a ChatGPT subscription is the other thing on your shortlist, the two vendors' individual tiers sit close enough on price that the work you do decides it.
• 2 to 150 people on Claude, no audit obligation. Team. You already get single sign-on, domain verification, JIT provisioning and spend caps, with usage included in the seat. The Claude Team plan for a small business covers the seat math.
• More than 150 people on Claude, or under an audit obligation. Enterprise. The trigger is either the Team seat cap or one named control from the Enterprise-only list, usually SCIM or audit logs.
• Machine workloads with no human behind them. The API. A nightly batch job does not need a seat, and giving it one wastes the fee.

When Enterprise beats putting the API on a corporate account

Enterprise usage is billed at standard API rates. So moving a team from a corporate API account to Enterprise does not change what a token costs unless you enable US-only inference. It adds a fixed fee per purchased seat, and that fee buys identity and audit controls a raw API key does not have.

An API-only setup gives you metered usage with no seat minimum and no monthly floor. No floor is not the same as free, though: the API has no standing free tier. New users get a small credit to test it, and as of October 2026 researchers working on qualifying AI safety and alignment topics can apply to Anthropic's External Researcher Access Program for $1,000 in API credits. Outside those grants the meter runs on every call. Max and Team subscriptions now include monthly API credits (as of October 2026, up to $500 a month pooled on Team), but those are a subscription benefit rather than a free tier, and Enterprise plans are not eligible for them. An API-only setup has its own identity and audit controls: Console supports SSO when enabled through Sales and SCIM for eligible organisations, and standalone Console organisations can query the Compliance API's Activity Feed. API-key workloads expose no message content through that API. Personal API keys identify a user; shared workloads can use service accounts or Workload Identity Federation.

The break-even is therefore decided by how many humans need supervised access. Multiply purchased seat count by the monthly seat fee, then compare that against building the same controls yourself: a proxy in front of the API, your own key rotation, your own logging, your own per-user attribution. Under 20 people the comparison starts with the cost of 20 purchased seats, including any you leave unused. At a few hundred people, the seat fee is usually smaller than the engineering it replaces. If those humans are engineers on Claude Code, the seat versus token comparison has its own numbers, and what a Claude Code seat includes on Team and Enterprise works through them. Run the same arithmetic at the individual level first with the break-even between the Claude API and a subscription, because the shape of the answer is the same and the numbers are easier to check.

FAQ

How much does the Claude Enterprise plan cost per user?

The published seat price is $20 per user per month, billed annually, as of October 2026. That figure covers the seat only. Every token your team spends is billed separately at standard API rates, or 1.1 times those rates for Opus 4.6, Sonnet 4.6 and later models if you turn on US-only inference, with no allowance included, so the total is seat price times purchased seat count plus actual consumption. No all-in price is published because usage per seat varies far too much to quote in advance.

What is the minimum number of seats for Claude Enterprise?

Twenty seats if you buy self-serve with a card or ACH transfer. Fifty seats if you buy through a salesperson. These minimums apply to purchased seats: a team with fewer than twenty people can buy twenty seats and leave some unused. Team itself caps at 150 seats, so an organisation that needs more than 150 seats moves to Enterprise whether or not it needs the extra controls. That cap counts the seats you buy, not your company's total headcount.

Does Claude Enterprise include usage, or is it billed separately?

Separately. Enterprise charges every token at standard API rates on top of the seat fee, and no allowance is bundled in. Pro, Max and Team include usage with the seat, bounded by usage limits, with optional paid usage credits beyond those limits. Self-serve Enterprise draws usage from credits bought upfront, while sales-assisted Enterprise is invoiced monthly in arrears.

Do I need Enterprise to get SSO for Claude?

No. Team already includes single sign-on, domain verification and just-in-time provisioning. What Team lacks is SCIM provisioning and audit logs, and those two are what usually force the move, along with the Compliance API that reads them back. If the requirement is that employees sign in with their work identity, Team meets it.

Is company data used to train Claude on the Enterprise plan?

No. Content on commercial plans is not used to train Anthropic's models by default, and that applies to Team as much as Enterprise. Enterprise adds custom data retention controls, a Compliance API for exporting activity logs, chat histories and file content, customer-managed encryption keys and a US-only inference option.