Claude Enterprise pricing: what you actually pay

Anthropic publishes the number most vendors save for the sales call: the Claude Team plan stops at 150 seats. Seat 151 is where self-serve pricing ends and procurement begins. A company of 200 to 2,000 people is past that line before the evaluation starts, so the real question is not whether to talk to sales but what the contract looks like when you do.

There are three ways to buy Claude for a company. Team seats: a published per-seat subscription for two to 150 people. The Enterprise plan: a seat fee plus usage at API rates, with a self-serve entry price and sales-assisted contracts above it. And the API itself: per-token consumption with no seats at all.

Every number below carries a date and a source. The structure is what survives the next price change.

How much does the Claude Team plan cost?

Standard seats are $20 per user per month billed annually, or $25 billed monthly; premium seats are $100 and $125. These are the only company-facing Claude prices Anthropic publishes as numbers, so they anchor everything else in this article.

Last checked: August 2026. Anthropic revises pricing; verify on the linked pages before you budget.

The model lineup refreshed with the Claude 5 family: Claude Opus 5 is a drop-in at the same $5/$25 rates Claude Opus 4.8 charged, and Claude Fable 5 opened a new top tier at double those rates. The structure of the bill did not move — which is the point of anchoring on shape rather than any single model’s price.

Both seat types include Claude Code — how to govern it at company scale is its own article. The difference between the seats is headroom: Anthropic’s support docs put a standard seat at 1.25x the Pro plan’s per-session usage and a premium seat at 6.25x, so one premium seat is roughly five standard seats’ worth of capacity for your heaviest users. Admins mix the two types freely, and can enable usage credits so people keep working after they exhaust the included limits. That option matters when you model cost: with credits enabled, the listed seat price is where spend starts, not where it stops.

The plan starts at two seats and stops at 150. Below that ceiling, if what you need is governed chat plus Claude Code for a handful of developers, the Team plan is the sensible buy; the contract math below starts to matter once you can see seat 151 from where you stand.

What does Anthropic publish about Claude Enterprise pricing?

The shape, and since mid-2026 part of the number. The pricing page states the model outright: “seat price + usage at API rates,” and now lists a self-serve Enterprise entry at $20 per seat with usage metered on top; larger sales-assisted contracts keep the same two dimensions with the seat fee negotiated. That shape deserves more attention than it gets: the contract has a fixed axis you sign for and a consumption axis that meters on top of it. Publishing the shape and a self-serve entry is more candor than most enterprise AI vendors manage.

The feature list is public, and it is substantive. Per Anthropic’s Enterprise page: SSO and domain capture, SCIM provisioning, role-based access control, spend controls, data retention controls plus audit logs with OpenTelemetry monitoring (both flagged as Enterprise-only), a Compliance API, IP allowlisting, and a HIPAA-ready option. Claude Code and company connectors are included. If your requirement is a managed assistant with a mature admin plane, and your security posture allows prompts and history to live in a vendor’s cloud, this is a real product for that job.

For the sales-assisted path, the math to run before the call is a pattern, not a spreadsheet you can fill in today:

annual cost = (provisioned seats × negotiated seat price × 12) + (a year of token usage at API rates)

The first term is fixed the day you sign. The second scales with behavior you cannot predict yet, because the usage data does not exist inside your organization until people start working. So arrive with questions that pin the second term down: which API rates apply and whether they are the public list rates, whether usage is committed upfront or pay-as-you-go, what the alerting looks like before spend surprises anyone, and what renewal does to the seat price.

Then hold the seat term up against how usage actually distributes. On the platforms we operate, consumption concentrates: a minority of people generate most of the tokens, and a long tail signs in occasionally. Per-seat billing charges both groups the same access fee. Anthropic’s hybrid model softens this compared with pure per-seat pricing, since light users burn little on the usage axis, but every provisioned seat still pays for access whether it works daily or quarterly.

An Enterprise quote has two axes: seats you provision and tokens you burn. Budget the second like a utility, not a license.

Is the Claude API cheaper than Enterprise seats?

For concentrated usage, usually. For broad, light usage, usually not. The API is the third way to buy Claude and it has no seats at all: you pay per token at the published rates in the table above, whether the tokens come from a chat interface, an IDE plugin, or an overnight agent run. When most of your volume comes from developers, scripts, and agents, per-token pricing tracks reality closely because idle people cost nothing.

The trade is governance. A subscription seat caps its own damage; the worst case is the price of the seat. An API key has no such reflex. Anthropic’s Console offers account-level limits, but per-person and per-team enforcement is yours to build, and without it the first monitoring system many companies deploy is the month-end invoice. A single misconfigured retry loop can spend a developer’s monthly seat-equivalent in an afternoon. This is the problem an LLM gateway exists to solve: virtual keys, per-key budgets, and spend attribution in front of the provider.

Which fork fits a 200–2,000-person company?

Usually more than one at once, which is why the decision feels messy. A company this size contains two populations: a broad group that needs governed chat a few times a week, and a narrow group of builders whose agents and coding tools would blow through any seat’s included usage. Anthropic’s own Enterprise pricing acknowledges this by billing both axes. The practical question is not which of the three paths to take but how much of each you sign up for, and which of the two numbers you can still control after signing. The seat count is a contract term. The token bill is an operational discipline.

Where SecondStack fits

SecondStack does not replace the Anthropic relationship. You bring your own Anthropic keys, or Bedrock and Vertex AI credentials, and the models bill to accounts you already control at the API rates above. What the platform adds is the layer that makes company-wide consumption workable: chat and agents for every employee, virtual keys for IDEs and scripts, and per-user, per-team, and per-key budgets with hard cutoffs. When a key reaches its budget, the gateway blocks the request before it leaves your infrastructure, so overspend stops at the source instead of surfacing on an invoice. There is no per-seat licensing; cost scales with what people actually use. The gateway itself is a hardened LiteLLM fork, and if you are weighing vanilla LiteLLM against the platform for this job, we have written up that comparison honestly.

The boundary is worth stating plainly: under roughly 150 people with straightforward chat needs, Anthropic’s own plans are the better buy. They are priced fairly for that segment and carry zero operational load. Past that size, the question stops being which plan and becomes which spend model you can still steer twelve months in. Price the seats, price the tokens, and sign the contract whose second year you can predict.