What does Claude Enterprise cost? Seats, usage, API rates and rollout services

The Enterprise plan of Claude, powered by Anthropic, costs $20 per seat per month, billed annually with a 20-seat minimum, plus every token your people use billed at standard API rates, as of October 2026. A 200-person company pays $48,000 a year in seats before any usage. Usage depends on which models and surfaces people use, and admins cap it with organization, group and user spend limits. Implementation services are a separate, one-time cost.

What do you pay for seats?

Anthropic prices Enterprise as a single seat type. The seat gives a person access to Claude on web, desktop and mobile, plus Claude Code and Cowork. It includes no usage. Seats are charged for the full year at the start of the term. Seats added mid-term are charged pro rata right away, and self-serve plans cannot remove seats mid-term, so do not buy for next year's headcount on day one.

How is usage priced?

Usage is charged per million tokens (MTok) at the same rates as the Claude API. Input is what Claude reads: your prompt, attached files, conversation history and tool results. Output is what Claude writes. Output costs five times input on every current model. Cached input, meaning context Claude has already read recently, is billed at a small fraction of the normal input rate.

Two modifiers matter for Enterprise budgets. First, if you turn on US-only inference, usage on Claude Opus 4.6, Claude Sonnet 4.6 and later models is billed at 1.1 times standard rates; seat fees do not change. Second, Anthropic sorts Claude surfaces by token intensity: core chat is lower, while Claude Code and Cowork are higher because each session carries system prompts, file context, tool calls and multi-step reasoning. Claude in Chrome runs as a Cowork session. Claude Tag channel work in Slack bills to the organization rather than a seat, so per-user and group limits do not cap it; it has its own spend limit.

How much usage should you budget per person?

There is no published per-user average, and anyone quoting one without your data is guessing. What you can do is build a model from assumptions, then replace the assumptions with real numbers after a pilot. The example below uses three usage profiles for a 200-person company. Every volume in it is an assumption we chose for illustration.

In this example, usage roughly equals the seat bill, so the all-in cost lands near $96,600 a year, or about $40 per person per month. Change one assumption and the total moves a lot. Ten more engineers add about $18,700 a year in usage alone. Moving the 40 analysts from Claude Opus 5.5 to Claude Sonnet 5.5 at the same volume cuts their usage in half. Turning on US-only inference adds about 10 percent to usage on current models.

How do you replace the assumptions with real numbers?

Run a 30-day paid pilot on Enterprise itself, not on personal Pro accounts, because only Enterprise shows you metered spend per person. Keep it small and deliberate.

• Pick 25 to 40 people across the three profiles above, including some skeptics, not only volunteers.
• Put each profile in its own group with a per-user spend limit, so one runaway session cannot distort the result.
• Set the organization default model on purpose, and record which model each group was told to use.
• Record month-to-date spend per member from Organization settings, then Usage, at the end of each week.
• Multiply each profile's median spend, not the mean, by the headcount you expect in that profile.
• Add headroom for the first year. We plan 20 to 30 percent and revisit after one quarter.

For comparison on the same 200 people: the Microsoft 365 Copilot enterprise add-on lists at $30 per user per month paid yearly, or $72,000 a year, on top of the qualifying Microsoft 365 licenses it requires. OpenAI does not publish ChatGPT Enterprise pricing. Claude Team is not an option at this size because it stops at 150 seats.

How do you keep the bill predictable?

• Start with group limits, not an organization cap. Anthropic recommends group-level and per-user limits first, because hitting the organization limit cuts off everyone at once.
• Set a per-user limit for each group, then override for individuals who need more. Individual limits always beat group limits.
• Use pooled group budgets (beta) for departments with a fixed budget and uneven use. Billing admins get email alerts at 50, 75, 95 and 100 percent.
• Set the organization default model. New conversations start on it, so it shapes most of the spend.
• Gate Claude Code and Cowork by group with custom roles, and roll them out in waves.
• Tell users that higher effort levels and heavier models cost more, and when to use each.
• On self-serve plans, watch the credit balance. At zero, usage stops for the whole organization. Auto-reload needs a card; it is not available with ACH.
• Review Time at limit on the Usage page monthly. People always at their limit may be your best adopters, so check what they produce before you cut them.

What does it cost to build custom agents on the API?

Agents your developers build on the Claude Platform run through a Claude Console organization; confirm with Anthropic how that usage is billed alongside your Enterprise contract. Token rates are the ones in the table above. On the API, batch processing cuts costs by 50 percent for work that can wait, and fast mode for Claude Opus 5.5 costs twice the standard rate. Platform tools carry their own charges, listed below. Monthly API credits come with Team and Max plans, not Enterprise, so budget API work as its own line.

What does a Claude rollout cost in services?

Licenses are only part of the cost. Someone has to configure identity, groups, limits and retention; connect company data; write the usage policy; and train people. You can do it with internal staff or hire a partner. Abstrakt Solutions consulting runs $180 to $225 per hour. We usually quote rollouts as a fixed fee or in phases after a discovery session, because scope depends on your identity setup, data sources and compliance needs. The table below lists the work a rollout usually includes, so you can compare a partner quote with what your own team would have to do.

Budget for internal time too. Plan on an owner from IT for identity and settings, a security reviewer for connectors and retention, and a business lead per department who decides what good use looks like. Skipping those roles costs more than any consulting line, because unmanaged usage and stalled adoption both show up on the invoice.

Where do Claude budgets go wrong?

• Budgeting seats only. On Enterprise, the seat is the smaller line for any team with heavy users.
• Using a consumer price as the benchmark. Pro and Max prices do not reflect metered Enterprise usage.
• Leaving the org default on the most expensive model for everyone, including light users.
• Buying seats for future headcount on a self-serve plan, where seats cannot be removed mid-term.
• Forgetting Slack channel usage. Claude Tag bills to the organization and needs its own limit.