Research Note · Anthropic · AI Pricing

Claude Enterprise pricing 2026: seat, API & compute cost reference.

The complete Anthropic Claude Enterprise pricing reference for enterprise buyers: per-seat tiers, per-token API rates across Opus 4, Sonnet 4 and Haiku 4, the Computer Use Tool and Projects consumption layer, the Bedrock versus Vertex versus direct decision, realised discount bands, and the negotiation levers that move Anthropic in 2026.

By James Hill-WoodUpdated Dec 202311 min readAI pricing research cluster
Bottom line

Claude Enterprise lists at $60 per seat per month on annual commit; realised pricing lands at $42–$54 per seat above 1,000 seats on a three-year term. A 1,000-seat rollout with moderate API usage comes in at $620K–$800K per year all-in. The hidden spend is consumption — Computer Use Tool and Projects — and the highest-value lever is a documented ChatGPT Enterprise alternative in hand.

01 Key findings

  1. Seat price is only one of four cost layers. Enterprise seats, per-token API consumption, the Computer Use Tool, and the Projects context pool each bill separately — the seat line understates true all-in cost by 15–40%.

  2. Enterprise is the only tier that clears security review. SAML SSO, SCIM, configurable retention, the zero-retention default and the contractual IP indemnity are Enterprise-only — the $35/seat delta over Team buys the controls, not more model access.

  3. Realised pricing runs well under list. $60 list becomes $42–$54 at 1,000+ seats on three-year commit; 15,000+ seat estates reach $24–$34 on three-year terms.

  4. Consumption is the fastest-growing, least-controlled spend. Computer Use runs $0.15–$0.85 per completed task and Projects overage is negotiated per-tenant — both need admin-console budget caps before rollout.

  5. Channel routing changes the accounting, not the token price. Claude on Bedrock or Vertex burns existing AWS EDP or GCP commit at identical token rates — a material win for hyperscaler-committed estates.

  6. Anthropic flexes on a credible alternative. Discount realisation tracks a documented ChatGPT Enterprise or M365 Copilot quote at equivalent seat counts more than any other single factor.

02 Claude Enterprise 2026 list pricing

Anthropic publishes Claude seat pricing across four named tiers in 2026: Free, Pro ($20/user/mo), Team ($25/user/mo annual, $30 monthly), and Enterprise (custom, with $60 per seat per month as the typical published reference). Enterprise is the only tier that includes SAML SSO, SCIM provisioning, custom retention windows, the zero-retention commitment, and the indemnity language enterprise buyers require. Claude Pro for individuals sits at $20/month.

TierList per user / monthContext windowModels available
Claude Free$0200K tokensClaude Sonnet 4
Claude Pro$20200K tokensSonnet 4 + Opus 4 (limited)
Claude Team (annual)$25200K tokensSonnet 4 + Opus 4
Claude Team (monthly)$30200K tokensSonnet 4 + Opus 4
Claude Enterprise$60 (typical, custom)500K tokens (expanded)All models incl. Haiku 4

A 1,000-seat Enterprise rollout at $60 lists at $720,000 per year. Realised pricing on a three-year commit at 1,000 seats typically lands between $42 and $54 per seat, or $504,000 to $648,000 per year. Discount realisation depends on TCV, term length, and the credible threat of ChatGPT Enterprise or Microsoft 365 Copilot in the deal.

Why the 500K window matters

Enterprise extends context from 200K to 500K tokens. For most knowledge work the 200K window suffices; for document-heavy workflows (legal review, contract analysis, large-codebase navigation, multi-document research) the 500K window is a step-change that often justifies Enterprise on its own. It holds roughly 1,200 pages of dense text — a full 10-K plus the prior year's filing and analyst transcripts in one context.

03 Enterprise versus Team versus Pro versus Free

The tier delta on a per-seat basis is $35/month — materially less than the comparable ChatGPT Business-to-Enterprise gap on most negotiated outcomes. The most common rollout pattern is Team for low-risk broad-population users (sales, marketing, general office) and Enterprise for high-risk knowledge work (legal, finance, research, engineering).

Pro
$20 / user / month
Best for: individuals and pilots — not production enterprise deployment.
Includes
  • Sonnet 4 + limited Opus 4 access
  • 200K token context window
  • Personal Projects
Excludes
  • No SSO, SCIM or admin audit log
  • No zero-retention default
  • No Computer Use, HIPAA BAA or SLA
Team
$25 annual / $30 monthly
Best for: low-risk broad-population users where enterprise controls are not mandatory.
Includes
  • Sonnet 4 + Opus 4
  • Team-level shared Projects
  • Basic admin console + MCP integrations
Excludes
  • No SAML SSO or SCIM
  • Limited retention control, no zero-retention
  • No Computer Use, HIPAA BAA or priority SLA
Enterprise
$60 typical (custom)
Best for: production deployment into security-, legal- and compliance-governed teams.
Includes
  • SAML SSO + SCIM provisioning
  • 500K context, org-level Projects with RBAC
  • Computer Use, HIPAA BAA, priority SLA
Notes
  • Zero-retention default; full audit log + export
  • Consumption billed on top of seat price
FeatureProTeamEnterprise
SAML SSONoNoYes
SCIM provisioningNoNoYes
Custom data retentionNoLimitedYes (configurable)
Zero-retention defaultNoNoYes
Admin console + audit logNoBasicFull
Context window200K200K500K
Projects (shared context)Personal onlyTeam-levelOrg-level with access controls
Computer Use ToolNoNoYes (consumption-billed)
Custom integrations (MCP)LimitedYesYes (priority support)
HIPAA BAANoNoYes
Priority support SLANoNoYes

04 Anthropic API per-token pricing

Most Claude Enterprise customers also consume the Anthropic API for embedded AI features in custom applications. API pricing is per-token and runs separately from seat pricing. The 2026 token prices for the production Claude 4 model family:

ModelInput / 1M tokensOutput / 1M tokensContext window
Claude Opus 4$15.00$75.00500K
Claude Sonnet 4$3.00$15.00500K
Claude Haiku 4$0.25$1.25200K
Claude Opus 4 (Batch)$7.50$37.50500K
Claude Sonnet 4 (Batch)$1.50$7.50500K
Cached input (1-hour)10% of standardn/an/a
Cached input (5-minute write)125% of standardn/an/a

Prompt caching is the largest single optimisation on the API. Workloads with stable system prompts (RAG, chat applications, agent frameworks) typically achieve 60–85% cost reduction on cached input tokens. Batch API pricing at 50% of standard applies to non-real-time workloads (overnight document processing, periodic embedding refresh, evaluation runs).

05 Projects & Computer Use consumption

Projects lets a team share context, instructions and knowledge files across users. It is included in Team and Enterprise tiers with consumption limits on the shared context pool; Enterprise adds organisation-level Projects with role-based access controls and audit logging that Team does not. The pool is bounded by the tenant's allocated capacity — heavy use cases (full corporate knowledge bases, large code repositories, multi-thousand-document research collections) can consume it faster than expected, triggering overage billing or expansion conversations. Overage pricing is negotiated per-tenant rather than published.

The Computer Use Tool lets Claude operate a virtual computer interface (screenshots, mouse, keyboard) to execute multi-step tasks in browser and desktop applications. It is billed per token (same Opus or Sonnet rates) plus a per-action overhead for the screenshot and execution loop. Typical agent runs in 2026 land between $0.15 and $0.85 per completed task depending on complexity and model variant. The justifying use cases are narrow but high-value: data entry into legacy systems without APIs, multi-step research and synthesis, and vendor-portal interaction.

Consumption is the seat floor's blind spot

The risk is unbounded agent activity running up large token bills against vague exploratory prompts. The mitigation is per-team budget caps and run-time limits enforced through the admin console — set them before rollout, because Computer Use is the highest-velocity new spend category in 2026 Anthropic enterprise contracts.

06 Claude on Bedrock versus Vertex versus direct

Claude models are available through three channels: direct Anthropic API, AWS Bedrock, and Google Vertex AI. Token economics are identical across channels — Anthropic does not pass discount through cloud sub-channels. The procurement decision turns on contract structure, deployment integration, and indemnity terms.

ChannelContract vehicleData residencyBest for
Anthropic directAnthropic enterprise MSAUS, EU expansion in 2026Anthropic-first organisations, full feature access
AWS BedrockBurns AWS EDP commitPer AWS regionAWS-committed estates, regional flexibility
Google Vertex AIBurns Google Cloud commitPer GCP regionGCP-committed estates, Gemini comparison workloads

For AWS-committed customers with material EDP burn, Claude on Bedrock is the right answer because token spend reduces the EDP commitment to be earned out — see AWS Bedrock pricing 2026 and the AWS EDP pillar for the burn-down economics. GCP-committed customers get the equivalent through Vertex AI. Organisations without material hyperscaler commits are typically simpler on direct Anthropic.

The MCP advantage

Claude was the first frontier model to natively support the Model Context Protocol (MCP), which standardises how the model connects to external tools, data sources and applications. Buyers integrating Claude with existing systems (Slack, Jira, Salesforce, internal databases) gain real integration velocity from the MCP ecosystem. MCP is an open standard, so it is not a permanent moat — but it is a genuine 2026 advantage when scoping enterprise rollouts.

07 Security, SSO & IP indemnity

The Enterprise tier includes the controls production security and legal teams require: SAML 2.0 SSO with major IdPs (Okta, Entra ID, Ping, OneLogin), SCIM 2.0 automated provisioning and deprovisioning, configurable retention from one day to indefinite, default zero-retention for prompts and completions (the model does not train on customer data), a full admin console with usage analytics, audit log with API export, HIPAA BAA availability, SOC 2 Type II, ISO 27001, and the Anthropic Acceptable Use Policy with named carve-outs.

Anthropic publishes a copyright indemnity for Claude output as part of the standard Enterprise MSA, covering direct copyright infringement claims from Claude-generated output, with standard exclusions for misuse, fine-tuning on infringing data, and prompts designed to elicit infringing content. It is structured similarly to the Microsoft and OpenAI commercial-output indemnities.

Data & IP terms — read the caps

The procurement question is how indemnity caps and notice provisions compare across vendors. Anthropic, OpenAI and Microsoft have meaningfully different cap structures and customer-side obligations that change the practical value of the indemnity. The cross-vendor comparison is documented in our AI contract data residency and IP rights guide.

08 Realised discount bands by seat count

Claude Enterprise discount realisation in 2026 follows a similar shape to ChatGPT Enterprise, with Anthropic typically slightly more flexible at higher-seat bands as it expands its enterprise sales motion. Realised three-year per-seat pricing falls sharply with scale:

250–500
~$52
500–1,500
~$47
1,500–5,000
~$41
5,000–15,000
~$35
15,000+
~$29
Seat count bandAnnual commit priceThree-year commit price
250 to 500 seats$55 to $62 per seat / month$48 to $56
500 to 1,500 seats$48 to $58$42 to $52
1,500 to 5,000 seats$42 to $52$36 to $46
5,000 to 15,000 seats$36 to $46$30 to $40
15,000+ seats$30 to $40$24 to $34

09 1,000-seat and 10,000-seat rollout cost model

All-in annual cost with seats, embedded API consumption, Computer Use at a 10% attach rate, and typical Projects overage:

Cost line1,000-seat rollout10,000-seat rollout
Enterprise seat licences (3yr commit)$540,000$3,600,000
API consumption (embedded apps)$60,000 to $200,000$500,000 to $1,800,000
Computer Use Tool (10% attach, moderate)$15,000 to $40,000$150,000 to $400,000
Projects context pool overage (typical)$5,000 to $20,000$50,000 to $200,000
Total annual$620K to $800K$4.3M to $6.0M

10 The buy framework

Four considerations drive a right-sized Claude Enterprise commitment. Weight them to your situation before signing.

Factor 01

Tier mix by risk

Deploy Team to low-risk broad-population users and reserve Enterprise for high-risk knowledge work. The $35/seat delta buys controls, not model access — do not over-license the whole population to Enterprise.

Factor 02

Channel routing

Material AWS EDP or GCP commit favours Bedrock or Vertex routing at identical token price. Validate that the commercial accounting benefit outweighs any feature or residency constraint.

Factor 03

Consumption exposure

Model API, Computer Use and Projects consumption before committing. Enforce per-team budget caps and run-time limits through the admin console so agentic spend stays bounded.

Factor 04

Alternative-vendor pressure

A documented ChatGPT Enterprise or M365 Copilot quote at equivalent seats is the single biggest determinant of Anthropic's flexibility. Run it in parallel, not after.

11 The five negotiation levers that work in 2026

These are the moves that move Anthropic. The first is decisive; the rest compound it.

Negotiate on price
When you hold a rival quote

Anthropic's flexibility correlates directly with documented OpenAI pricing at equivalent seat counts. Bring a ChatGPT Enterprise quote and a three-year term for meaningful unit-price relief.

Negotiate on channel
When you hold cloud commit

With material AWS EDP or GCP commit, route via Bedrock or Vertex to consume from the existing commitment. Token price is identical but the commercial accounting changes materially in your favour.

Negotiate on structure
When forecasts are uncertain

Commit at 65–75% of forecast initial seats with quarterly expansion at the contracted unit price, and bundle API consumption alongside seats — account teams can discount embedded-app usage for seat commit.

Parallel process Recommended

Run Anthropic and the ChatGPT Enterprise / M365 Copilot alternative concurrently, with each aware a primary decision is live. This creates the competitive tension that unlocks the best unit price and consumption terms.

Sequential process Weaker

Negotiating Anthropic first, then shopping the outcome, relaxes pressure and lets the incumbent anchor the deal. Total leverage — and realised discount — falls.

The full Anthropic procurement framework lives in our enterprise AI vendor selection framework. For cross-vendor token comparison see the enterprise LLM cost comparison; for the per-token negotiation pattern see AI usage-based pricing negotiation. For broader counsel see AI procurement advisory, the AI procurement guide, software licensing advisory, and cloud contract negotiation.

Negotiate Claude Enterprise with the full picture

Independent Claude Enterprise reviews consistently land a median 22% below Anthropic's opening quote — in shadow advisory or on your behalf.

Request AI procurement advisory →

The Licensing Edge

Weekly vendor and AI-pricing intelligence for enterprise IT leaders. 3,000+ subscribers.