Claude Enterprise Pricing (2026): The $20 Seat Anthropic Now Publishes, the Usage Bill on Top, and Who Actually Needs It

TL;DR: As of September 2026, Claude Enterprise pricing is published: $20 per seat per month, billed annually, plus usage billed at API rates, straight from Anthropic's pricing page. That replaces a year of "contact sales" and the $60-to-$150-a-seat guesses that filled the gap, including on this site. The seat is the small number. The usage line is the one you have to estimate, and this page shows how, with Anthropic's own per-token rates. The short version for most readers: if your company is under 150 people and no regulator is making you buy audit logs, the plan you want is Claude Team at the same $20 seat with usage bundled in.

What Claude Enterprise costs in 2026

From claude.com/pricing on September 9, 2026:

• Seat price: $20 per seat per month. Enterprise is only sold on annual billing; Team can be monthly or annual.
• Usage: billed at API rates on top of the seat. Anthropic's wording is "usage cost scales with model and task." Team, by contrast, bundles usage into the seat and sells a $100-a-month premium seat with five times the allowance.
• Two ways to buy it: Enterprise (self-serve), bought with a card from the pricing page, and Enterprise (sales-assisted), through a conversation with Anthropic's sales team.
• What the seat adds over Team, in Anthropic's list: admin-set spend limits for users and the org, role-based access, SCIM provisioning, audit logs, a Compliance API, custom data retention controls, network-level access control, IP allowlisting, a HIPAA-ready offering, and Claude Security in beta.
• Models and context: Anthropic's comparison table gives Enterprise a 500k context window on the default model against Team's 200k, and marks the Fable model as available on Enterprise and not on Team. Opus, Sonnet, and Haiku are on both.

The full plan map, free through Enterprise, lives on the Claude AI pricing page. This one is about the Enterprise line specifically, because the number just changed and most of the internet still has the old one.

How much is Claude Enterprise?

$20 per seat per month, billed annually, plus usage at API rates, with a 20-seat minimum on a self-serve purchase and 50 on a sales-assisted one (Anthropic's pricing page and Enterprise support article, re-read October 12, 2026). So the smallest possible Enterprise bill is $4,800 a year before anyone sends a message. The table puts the seat floor next to an all-in estimate, using the seat profiles in the next section: about $275 a month of usage for a heavy Opus seat and about $2 for a light Sonnet seat, with roughly 15% of people heavy. The Team column is the same people on Team, standard seats at $20 and the heavy users on $100 premium seats, usage included.

Those usage figures are illustrations built from Anthropic's published token rates, not quotes. The point they make holds at every size: on Enterprise the seat is the small number, and the admin spend limits exist because the meter is the big one. Team stops at 150 seats, so past that the comparison is Enterprise or nothing.

The usage bill: the part you have to estimate

"$20 a seat" is not the price. It is the floor. Every message, document, and Claude Code session on an Enterprise plan is metered at the same rates the API charges, which as of Anthropic's pricing docs are $2 per million input tokens and $10 per million output tokens on Sonnet 5, $5 and $25 on Opus 5, and $1 and $5 on Haiku 4.5. The per-model detail is on the Anthropic API pricing page. Here is what that does to a seat, as arithmetic on those published rates:

• A light seat, someone who asks a few questions a day and drafts the odd email: call it 500,000 input tokens and 100,000 output tokens a month on Sonnet 5. That is $1 in and $1 out. Roughly $22 a month all in.
• A working seat, someone who runs documents, spreadsheets, and long threads through it daily: 5 million in, 1 million out on Sonnet 5 is $10 plus $10. Roughly $40 a month.
• A heavy seat on Opus, say a developer in Claude Code most of the day: 30 million in, 5 million out is $150 plus $125. Roughly $295 a month, for one person, before prompt caching brings the input side down.

Those are illustrations, not quotes; your token counts will differ, and Anthropic notes that its 4.7-and-later models count roughly 30% more tokens for the same text. But the shape is the point. A twenty-person company where three people are heavy users and seventeen are light is paying $400 for seats (20 at $20) and $900 or so for usage, about $1,300 a month in all, and the usage side is what the admin spend limits exist to cap. On Team, the same twenty people pay $400 a month on annual billing and the three heavy users get premium seats at $100 each, for $640 flat, with no meter. That is the real comparison, and it is why "Enterprise is only $20" is the wrong takeaway.

Claude Team vs Claude Enterprise: the decision

The seat price is identical on annual billing, so the choice is about two things: whether you want usage bundled or metered, and whether you need the controls.

• Choose Team if you have 2 to 150 people, you want a predictable bill, and your requirements are SSO, central billing, admin control over connectors, and the no-training default. Team has all of those. Standard seats are $20 annual ($25 monthly); premium seats with five times the usage are $100 annual ($125 monthly).
• Choose Enterprise if a compliance or IT requirement names something on the Enterprise-only list: SCIM provisioning from your identity provider, audit logs, a Compliance API your security team will actually poll, custom data retention, IP allowlisting, domain capture, or the HIPAA-ready offering. Or if your usage is so lopsided that metering it beats buying premium seats for the few heavy users.
• Choose sales-assisted Enterprise if you need to pay by invoice on net terms, buy through AWS Marketplace against an existing commit, negotiate tiered incentives on committed spend, or get a trial. Anthropic's table marks all of those as sales-assisted only.

One thing the seat price does not tell you: Enterprise is annual-only. A company that wants to try Claude for a quarter is a Team customer by construction.

Can you buy Claude Enterprise self-serve?

Yes. The pricing page has a "Create plan" button for Enterprise (self-serve) beside the "Contact sales" one, and you pay by credit card or ACH on an annual term (pricing page, October 12, 2026). What you can't get without talking to sales: invoicing and net terms, buying through AWS Marketplace, a trial, a consultation, or committed-spend discounts. The 20-seat minimum applies either way, and the HIPAA-ready offering is on the self-serve column too.

Nonprofits get a different price sheet. Anthropic offers Team at $8 per user per month to verified nonprofits (against $20 annual), and Enterprise at nonprofit pricing through its sales team for organizations of 20 or more people; the eligibility rules and the math are on Claude for nonprofits.

Self-serve vs sales-assisted: what changes and what does not

Both flavors are $20 a seat plus usage, both are annual, and both get the security and admin feature set. Anthropic's comparison table separates them on commercial terms: credit card on both, ACH on both, invoicing and net terms on sales-assisted only, AWS Marketplace, a consultation, committed-spend incentives, trials, and customer-success support at spend thresholds on sales-assisted only.

The place this matters most is HIPAA. The pricing page lists a HIPAA-ready offering on both Enterprise columns and not on Team, and as of October 10, 2026 Anthropic's BAA documentation agrees: the Business Associate Agreement covers Enterprise plans, self-serve included, once the workspace's Primary Owner switches HIPAA on and accepts it in settings (it said sales-assisted only when this page launched). Features like Cowork in the cloud and the beta Office add-ins stay outside it even then; the detail is on the is Claude HIPAA compliant page. If patient data is anywhere near this decision, do not infer coverage from a checkout page. Ask for the BAA, get it in writing, and confirm who activated it.

What the old numbers got wrong

For most of 2025 and the first half of 2026, Anthropic sold Enterprise by quote only, and the internet filled the vacuum with ranges: $60 to $150 a seat, 20-to-50-seat minimums, five-figure annual floors. Those numbers came from reported deals and they may have been true for those deals. The per-seat prices are not what the pricing page says now, and I have corrected them on my own pages as I found them. The seat floor survived, though: the pricing page does not print it, but Anthropic's Enterprise support article lists 20 seats for a self-serve purchase and 50 for a sales-assisted one. If you are reading a comparison that still leads with "Enterprise is custom, expect $14,400 a year minimum," it was written before the page changed. The published shape today is a $20 seat, an API-rate meter, a self-serve checkout, and a 20-seat floor (so $4,800 a year before usage).

Claude Enterprise vs ChatGPT Enterprise on price

As of September 2026, Anthropic publishes the Enterprise number and OpenAI does not. OpenAI's pricing page lists ChatGPT Business at $20 per user per month billed annually ($25 monthly, two or more users) and sends Enterprise to a contact-sales form with no price. The control sets are near-identical on paper: SCIM, role-based access, audit logs through a compliance API, data-retention controls, no training on business data by default. So the practical comparison for a company under 150 people is Claude Team against ChatGPT Business, which cost the same, and the tie-breaker is which one sits inside the tools your company already runs.

If ChatGPT Enterprise is still on your shortlist, what the contract includes and how OpenAI bills it without a list price (seats on a credit-based agreement, or token-based billing on eligible agreements, invoiced monthly in arrears) is in ChatGPT Enterprise pricing.

Claude Enterprise vs Team

Side by side, from claude.com/pricing and Anthropic's Team plan and Enterprise plan support articles, all read on September 15, 2026.

Where Enterprise starts to make sense: at 20 seats it becomes possible to buy self-serve, and it pays off when a security requirement names something in the last row, or when usage is lopsided enough that metering a few heavy users beats buying them Premium seats. Under 20 people it is not on offer without a sales conversation, and Team is the answer.

The owner's answer

Most people who search "Claude Enterprise pricing" run a company of five to fifty people and typed the word because it is the top of the pricing page. If that is you: buy Team, put the heavy user on a premium seat, and spend the money you would have spent on Enterprise on the twenty-minute context document and the three workflows that make any plan pay for itself. That setup is what I install, and the productized versions of it are on the Claude for Small Business page: the Playbook at $397 if you want to do it yourself, the AI Readiness Assessment at $2,500 if you want the plan first, and the install from $4,500 if you want it done. If a regulator, an IT policy, or a real audit-log requirement is the reason you are on this page, Enterprise at $20 plus usage is a fair price for the controls, and the metered usage is the line to model before you sign the annual contract.

Update, October 21, 2026: what Claude Security costs on Enterprise

Anthropic made Claude Security continuous on October 21. Instead of waiting for someone to start a scan, it watches the repositories you connect as the code changes, marks a finding fixed when the fix ships, and reopens it if the bug comes back. It's in public beta for Claude Enterprise only (Anthropic's announcement).

The billing is the part the announcement skips. Per Anthropic's help center, you need an active Enterprise plan, a premium seat for each user, and usage credits turned on for consumption billing. Claude Security is charged at direct token cost with no platform fee, and that usage covers the scans, the threat model Claude writes for each project, the project chat and the fix checks. You can give it its own spend limit and see its cost per day in each project's Usage settings. The launch offer: eligible Enterprise organizations get up to $25,000 a month in free usage credits for two months, through December 31, 2026.

What that does to the math above: the $20 seat and the 20-seat minimum don't change, but continuous scanning adds a second metered line to the usage bill. Set its spend limit before the two free months run out, or the first full invoice is the first time you'll see what scanning your codebase every day actually costs.

Update, October 23, 2026: Cowork and chat merge on Enterprise, and four ways to cap the meter

On October 23 Anthropic made chat and Cowork one Claude for Enterprise plans, the same change Pro and Max got in September (Anthropic's announcement). A conversation starts as a chat. When a task needs to run code or work with files, Claude carries on in a cloud session that Anthropic runs, and the work keeps going after the laptop closes. Admins switch it on in Organization settings. Organizations on zero data retention for Claude Code, the HIPAA-ready setup or customer-managed keys stay on their current configuration.

The price doesn't change: Enterprise is still usage-based, and the announcement says agentic work "can use more tokens than a quick chat." That's the meter from the usage section above getting bigger. What's new is the set of controls for it:

• Efficient mode, a new setting that keeps the unified Claude on Sonnet and Haiku models.
• Model entitlements, which set which models each role can use.
• Spend limits at the organization, group and member level.
• Pooled group budgets, one shared monthly amount a group draws from.

Here's what the first one does to the heavy seat in the usage section. That developer's 30 million input and 5 million output tokens cost about $275 a month on Opus 5. At the Sonnet 5 rates on this page ($2 and $10 per million) the same tokens cost about $110. Anthropic also says Opus 5.5 runs about 40% cheaper than Opus 5 on typical workloads, so the gap narrows on the newest model. The order I'd set it up in: Efficient mode for everyone, Opus entitled only to the roles that need it, and a pooled budget per team, so the first big invoice isn't the first time you find out who's running agent work all day.

Update, October 28, 2026: pooled group budgets go GA, and how to size one

Pooled group budgets moved from the list above to generally available on October 28, alongside two reporting tools (Anthropic's announcement). The price still doesn't change. What the announcement adds is how the pool and the personal limits interact, and that decides how you set them:

• Both limits count. Every request is charged against the member's own limit and the group's pool, and the person stops at whichever runs out first.
• Asking for more happens inside Claude. A member who hits a limit sends a request from the app, and it goes to admins for review.
• Setup suggests a number from the group's recent usage, and admins with billing permission get emails at 50%, 75%, 95% and 100% of the budget.

Why the pool matters, in numbers. Take a 10-person team where two people do heavy agent work worth about $150 a month of usage each and the other eight use about $12 each, $396 in total. With a flat $40 limit per person ($400 of total headroom), the two heavy users hit their cap around day 8 while the light users leave about $224 unspent, so the team pays for headroom it can't use and the two people getting the most out of Claude are the ones blocked. Put the same $400 in a pool and raise the personal limits to $175, and the whole month fits, with the personal limit still catching a runaway session. These are illustrations, not quotes, but the shape holds: keep personal limits well above the average seat and let the pool be the cap you watch.

The other two tools are for proving the spend was worth it. Smart reports group a sample of a team's conversations into workstreams and show what each one costs, what it produced and where it got stuck (a missing connector, a failed tool call), and admins with billing permission can raise a limit straight from the report. In-app surveys ask members what Claude changed about their work. Anthropic says the reports aren't for judging individual performance. Availability varies for organizations on a HIPAA configuration or customer-managed encryption keys, so check before you plan around them.

Update, November 5, 2026: the security controls your IT team will ask about

On November 5 Anthropic added a batch of security and admin features to Claude Enterprise (Anthropic's announcement). The announcement lists no new price, so the $20 seat plus usage above still stands. What changes is the answer to the question that usually sends a company from Team to Enterprise in the first place: can our security people see and control what Claude does?

• Security center. A page in Organization settings that compares your Claude settings for chat, Cowork and Claude Code with Anthropic's recommended setup, shows the current and recommended value for each control, and links to where you change it. You can download it as a report for an auditor.
• Inference hooks, now generally available. Prompts, and tool calls before they run, get sent to the data security tool you already use, so the rules you wrote to stop customer data leaving the building apply to Claude too.
• One admin console at admin.claude.com for a company with several Claude organizations: single sign-on, automatic user setup from your directory, verified domains, groups and Compliance API keys, set once instead of per organization. More than one identity provider is allowed, which matters after an acquisition.
• Access controls. Sign a person out of every browser and device with one Admin API call, manage your own IP allowlist without going through Anthropic, and block personal Claude accounts in Claude Desktop on company computers.

Two limits the announcement states plainly, and both are worth knowing before you promise anything to an auditor. The security center "is guidance based on Anthropic's recommendations, not a certification or a guarantee of compliance with any law or standard." And the one-call sign-out ends sign-in sessions only: it doesn't end API keys, cloud sessions already running, or scheduled routines, and the person can sign back in. So an offboarding checklist still needs a step for each of those. If your company is under 20 people and nobody in IT has asked for these, nothing here changes the advice above: Team is still the plan.

Update, November 6, 2026: spend by client or cost code, and what it leaves out

A day later Anthropic added project tags to Claude Enterprise, in beta (Anthropic's help article). An owner creates a tag, like client_matter or cost_code, and each project gets one value for it, up to 16 tags a project. Organization settings > Usage > Spend by project then groups spend by that tag, with a CSV export and an API for your billing system. For a law firm or an agency that bills clients for AI time, this is the first built-in way to answer "what did Claude cost on this matter?"

Read the numbers knowing what they don't count, because the help article is plain about it:

• Seats aren't in it. Usage covered by seats isn't counted, so a project used mostly on seat allowance can show little or no spend. The report shows the usage meter, not the $20 seat.
• It isn't your invoice. Amounts are after your discount and before credits, and recent numbers can change until about a week after month end. Data runs through the end of yesterday (UTC).
• Only project work gets a tag. Chat and Cowork inside a project count. Claude Code counts only after each person puts the project ID in their Claude Code settings file; without it the usage shows as "Not in a project" and nobody gets a warning. Claude for Office, Claude in Chrome, Slack and Teams, voice mode, artifacts and the API don't get tagged at all.
• Tag first, then work. A new tag takes about two hours to take effect, earlier spend in that project never counts toward it, and changing a tag doesn't move past spend.
• Not every Enterprise org gets it. It's off for organizations on a HIPAA configuration, zero data retention, customer-managed encryption keys, or data stored outside the US. Tag names and values are visible to everyone in the organization, so a client's name in a tag is visible to every employee.

No new price came with it. If you need to bill AI back to clients, set the tags and the Claude Code settings line before the first matter starts, then reconcile the report against the invoice each month rather than sending the report as the bill.