Claude Enterprise Pricing in 2026: Seat Fees, Usage at API Rates and Private AI

Claude Enterprise pricing is US$20 per seat per month, billed annually, plus every token your organisation uses at Anthropic's API rates, with a minimum of 20 seats (verified 14 October 2026). The seat buys access; usage is billed on top. Claude Team, for 2 to 150 people, charges $20 per Standard seat billed annually or $25 monthly, and $100 or $125 for a Premium seat, with usage included up to limits.

Anthropic sells Claude to companies on two plans with different economics. Team is a fixed price per seat, with usage included until a member reaches a five-hour or weekly limit. Enterprise separates the two: the seat fee pays for access, and chat, Claude Code and Cowork usage is billed per token at the rates developers pay on the Claude Platform. Older Enterprise seat types that bundled an allowance are no longer offered on new contracts.

Everything below was checked on 14 October 2026 against Anthropic’s pricing page, its Enterprise page, its help centre and the Claude Platform pricing documentation. Our ChatGPT seat-cost breakdown covers OpenAI’s plans in the same format.

Claude Team and Enterprise list prices

Three details change the totals. Anthropic’s prices exclude tax. Seen from an EU address on the same day, the page quoted Team in euros, €18 per Standard seat billed annually or €21.04 monthly and €90 or €105.23 for Premium, while Enterprise stayed in US dollars; the help centre confirms that self-serve Enterprise is billed in US dollars only. And Team members who run out of allowance can keep working on usage credits, which Anthropic charges at standard API rates.

Claude Enterprise pricing per seat: what the $20 buys

Anthropic’s billing article is plain about it. The Enterprise seat gives one person access to Claude on the web, desktop and mobile apps, and every token the organisation consumes is billed separately at standard API rates. Seats carry no individual token allowance, and usage is billed as one organisation-wide total, whoever generated it.

How usage reaches the invoice depends on how you buy:

• Self-serve Enterprise. Usage is purchased up front as credits, which draw down as people use Claude and Claude Code. ACH payment is accepted; invoicing is not.
• Sales-assisted Enterprise. Usage is invoiced monthly in arrears on actual consumption, with net terms. This route also adds AWS Marketplace purchasing, trials and incentives on committed spend.

Seat rules matter for the budget. Seats added mid-term are charged pro rata for the rest of the year, immediately. On self-serve plans the seat count can only go down at renewal, and removing a member frees the seat for someone else without reducing the bill. Owners can cap spend per user and for the whole organisation, but they cannot switch usage billing off.

Claude Enterprise pricing per token

Model choice moves the bill more than anything else: the same request costs five times as much on Fable 5.1 as on Sonnet 5.5. Two multipliers sit on top. Pinning inference to the United States adds 10% to every token price on Claude 4.6 and later models, and Anthropic notes that the tokenizer used from Claude 4.7 onward produces roughly 30% more tokens for the same text than its predecessor.

Claude Team vs Enterprise pricing and controls

The comparison has a twist. A Standard Team seat on annual billing and an Enterprise seat cost the same $20, yet only the Team seat comes with usage. Moving to Enterprise buys administration and compliance features and removes the 150-seat ceiling; it does not lower the price per person, and for heavy users it can raise it. Anthropic’s Enterprise page also lists SOC 2, ISO 27001, GDPR and CCPA compliance and audit logs with OpenTelemetry monitoring. What each plan does with prompts and files is the subject of our Claude confidentiality guide.

How Claude Code is billed on each plan

On Team, Claude Code comes with every seat and draws on the same allowance as chat and Cowork, which resets on a rolling five-hour window and a weekly one. Premium seats are meant for the heavier users. When members hit their limits, owners can turn on usage credits at API rates and set monthly caps for the organisation and for each person. The plan’s pooled API credits, $20 per Standard seat and $100 per Premium seat up to $500 a month, cover the Claude API, the Console, Managed Agents and the Agent SDK. They do not cover interactive Claude Code or extra usage in the apps, and unused credits expire at the end of each billing cycle.

On Enterprise, Claude Code usage is billed at API rates with no per-seat limit. Anthropic’s own cost documentation supplies the planning figure: across enterprise deployments, Claude Code averages about $13 per developer per active day and $150 to $250 per developer per month, and 90% of users stay below $30 per active day. For a developer the $240 annual seat is therefore the small part; usage at those averages adds $1,800 to $3,000 a year.

Anthropic advises piloting with a small group to set a baseline, and the spend report in org analytics breaks estimated spend down by user and by model. Teams that want part of their coding work on open-weight models can follow our Claude Code local-model setup.

Claude Enterprise cost for 250 and 1,000 seats

The table uses list prices with annual billing. Two inputs are assumptions and are labelled as such: that one seat in five belongs to a developer using Claude Code, and that everyone else processes 1 million input tokens and 100,000 output tokens a month in chat. Replace both with pilot data before the numbers go into a budget.

How to read it:

• Usage is most of the bill. In this scenario the seat fees are 28% to 38% of the total. The share of developers and the model mix decide the rest.
• Residency costs extra. Pinning inference to the US multiplies the usage lines by 1.1.
• Spend limits cap usage; they do not include it. Enterprise has no allowance to fall back on.

Claude Team for companies of up to 150 people

A 120-person company on Team with 100 Standard and 20 Premium seats on annual billing pays 100 × $240 plus 20 × $1,200, or $48,000 a year, with usage included up to each member’s limits. Its seats would earn $4,000 of monthly API credits, so the $500 cap applies. The same 120 people on Enterprise would pay $28,800 in seat fees before a single token of usage.

What the Claude seat price leaves out

• Usage, on Enterprise. All of it, at API rates, prepaid as credits or invoiced in arrears.
• What your developers build. Applications on the Claude Platform are billed per token outside the seats. Team’s pooled credits offset up to $500 a month, and Claude used through Amazon Bedrock, Google Cloud or Microsoft Foundry is billed through that cloud rather than your Claude plan.
• Residency. US-only inference at 1.1 times standard rates.
• Extra usage on Team. Usage credits beyond each member’s allowance, at API rates.
• Seat commitments. Annual terms, pro rata additions and no mid-term reductions on self-serve Enterprise.
• Rollout. Identity set-up, a security review, an acceptable-use policy and staff training, whichever plan you choose.

If Microsoft or Google is already your main supplier, the same lines are worked through for Microsoft 365 Copilot and Gemini Enterprise.

Where a private deployment’s costs come from

Claude Enterprise turns AI work into a variable bill that grows with every prompt, file and tool call. A private deployment turns the same work into capacity you own or lease, and its cost lines look different:

• Accelerators sized for concurrency. The number of simultaneous requests at peak sets the GPU count, not headcount. Coding agents, which send whole files and loop through many steps, are what push that peak up.
• Power, cooling and space for as long as the servers run.
• Operations. Patching, monitoring, model upgrades and user support, usually as a share of an existing platform team’s time.
• Software. A platform licence if you buy one, and support for the serving stack and the open-weight models.
• Utilisation. Idle hardware costs as much as busy hardware, so the cost per request only falls as usage rises.

Open-weight models are not Claude, and whether they are good enough for a given team is a quality question to settle in a pilot before any spreadsheet. A hybrid is possible: keep a frontier cloud model for the hardest requests and serve routine ones locally. Our self-hosted route for chat and coding describes the switch, the breakeven workings cover hardware, power and staffing over several years, and the flat vs pay-as-you-go guide explains why the commitment model matters more than the billing unit.

How to compare the two on equal terms

1. Measure tokens, not seats. Run a pilot and export the spend report, which breaks spend down by user and model.
2. Split coding from chat. Price developers at Anthropic’s Claude Code averages, or better at your own pilot figures, and price chat users separately.
3. Fix the model mix. A Sonnet-only estimate and an Opus-heavy one can differ by a factor of two.
4. Size private capacity on peak concurrency, and count operations staff at a realistic share of their time.
5. Use one horizon. Set three years of seats and usage against three years of hardware, power, staff and software.
6. Take out what cannot leave. Data that may not go to a cloud processor at all sits outside a Claude price comparison.

Finance teams turning this into a budget line can start from our CFO guide to AI spend and test seat counts in the AI savings calculator.

How VDF AI fits

VDF AI is a self-hosted enterprise AI platform: chat, private RAG over company data, governed agents and workflows, audit logs and role-based access, plus VDF Code for on-premises coding assistance. Because it runs on your infrastructure, its economics follow the capacity lines above rather than a per-token bill.

On-premises, VDF AI is licensed by seat in Seat Bundles, with API Expansion Packs added when other systems call the platform programmatically, and the licence does not meter model tokens. VDF AI Router places Ollama endpoints and custom on-premises model deployments next to whichever cloud models your policy approves, and in air-gap mode it routes to local models only. On-premises list prices are provided on request; see current pricing for the options.