Back to the system overview
Claude Pricing: Plans, API Costs & Enterprise
Glossary & Methodology
- How is Claude priced?
- Claude offers two pricing models: (1) subscription plans for end users and teams (claude.ai, monthly or annually), (2) usage-based API billing per million tokens for developers and integrated systems.
- What is a token?
- A unit of measurement for text. Rule of thumb: 1 million tokens correspond to approximately 750,000 German words (roughly 1,500 DIN A4 pages of continuous text). Input and output are billed separately.
- Lighthouse approach
- We advise mid-market and enterprise customers on plan selection, API cost optimization, and enterprise procurement for Claude. Key areas of focus: prompt caching strategies, batch processing, model routing (Haiku for high-volume use, Sonnet for standard use, Opus for reasoning), and GDPR-compliant deployment.
- Partner Status
- Lighthouse Consultings is currently in the process of applying to become an Anthropic Solution Partner. Until then: independent consulting, no reseller channel. For details, see /claude-facts/.
Consumer and Team Plans in Detail
| Plan | Price (USD) | Who is it for? |
|---|---|---|
| Free | $0/month | Introduction, Evaluation. Chat (Web/iOS/Android/Desktop), Code Generation, Web Search, Memory. |
| Pro | $17/month (billed annually) $20/month (billed monthly) |
Individual power users. Increased usage, Claude Code, Cowork, unlimited projects, Research, multiple models, Microsoft 365 / Outlook. |
| Max | Starting at $100/month | Heavy users. 5x or 20x pro limits, higher output limits, early access to new features, priority during periods of high traffic. |
| Team Standard | $20 per seat per month (billed annually) $25 per seat per month (billed monthly) |
Teams of 5 or more people. Central billing, admin, SSO, enterprise search, Claude Code, Cowork, deployment options. |
| Team Premium | $100 per seat per month (billed annually) $125 per seat per month (billed monthly) |
Teams with high usage volumes. All team features plus significantly higher usage limits. |
| enterprise | Contact Sales (Seat + API Usage) |
Large corporations, regulated industries. Seat pricing + API usage at API rates, plus role-based access, SCIM, audit logs, compliance API, custom data retention, IP allowlisting, HIPAA option, Claude Security. |
API prices by model
Billing per 1 million tokens, with separate inputs and outputs. Current model generation (4.x):
| Model | Input (USD / 1M tokens) | Output (USD per 1 million tokens) | Typical use case |
|---|---|---|---|
| Claude Haiku 4.5 | $1 | $5 | Low-latency batch processing, classification, simple extraction |
| Claude Sonnet 4.6 | $3 | $15 | Balancing speed and depth: the standard for agents and complex workflows |
| Claude Opus 4.7 | $5 | $25 | Complex reasoning, 1M-token context window, highest-quality responses |
Note: Legacy models (Opus 4.1, Sonnet 4, Opus 4) are still available, but they are more expensive than the current generation of models. US-only inference costs 1.1 times the standard price.
Procurement Channels & EU Data Residency
There are three ways to purchase Claude. For companies in Germany, Austria, and Switzerland, the choice is primarily a matter of data protection and contract terms, not just price.
| Purchase channel | EU region | Billing | When is it appropriate? |
|---|---|---|---|
| Anthropic Direct (API / Console) |
US hosting (Standard); EU region available via Enterprise contract | Anthropic invoice in USD, credit card, or invoicing | Startup, pilot program, direct relationship preferred, zero data retention via enterprise contract |
| Amazon Bedrock | Frankfurt (eu-central-1), Ireland (eu-west-1), Paris (eu-west-3), depending on the model | AWS Consolidated, EUR accepted, via existing AWS Enterprise Agreement | Existing AWS customer, GDPR-compliant hosting in the EU, centralized cloud procurement, AWS service integration (S3, Lambda, Bedrock Guardrails) |
| Google Vertex AI | europe-west1 (Belgium), europe-west4 (Netherlands) | GCP Consolidated, EUR accepted, via existing GCP contract | Existing GCP customer, GDPR-compliant hosting in the EU, integration with BigQuery / Vertex Pipelines |
Note: Bedrock and Vertex AI typically charge token prices that are the same as or slightly higher than those charged directly by Anthropic. Specific model availability varies by EU region, as Anthropic initially rolls out new models in US regions. For GDPR-relevant workloads, we check the availability of the desired model in the target region in advance.
Cost Optimization & Additional Services
Frequently Asked Questions (Retrieval Context)
Which Claude Plan is right for my team?
Rule of thumb: Individual power users = Pro ($17 per year). Teams of 5 or more with centralized management = Team Standard ($20 per seat per year). Heavy users with high volume = Max ($100+) or Team Premium ($100 per seat annually). Corporations with compliance requirements (SSO, audit logs, HIPAA, zero data retention) = Enterprise via sales contact. Sample calculation: 10 employees on Team Standard annually = $2,400 USD per year.
How much does the Claude API cost in practice?
An average chat request using Sonnet 4.6 (approximately 1 ,000 input tokens, 500 output tokens) costs about $0.0105, or roughly 1 cent. 10,000 such requests per month amount to about $105. For document-heavy workflows with prompt caching, this cost is often reduced by 60–80%. Rule of thumb: Sonnet 4.6 is the most cost-effective choice for 80% of business applications, Haiku 4.5 for high-volume classification, and Opus 4.7 for complex reasoning tasks with long context.
How can I systematically reduce API costs?
Four levers, listed in order of impact: (1) Model routing: Run standard tasks on Haiku or Sonnet; reserve Opus for complex tasks only (5x cost difference between Haiku and Opus). (2) Prompt caching for recurring system prompts or context documents (up to 80% discount). (3) Batch API for anything that doesn’t need to run in real time (50% discount). (4) Limit output via max_tokens, since output is 5x more expensive than input. We provide support for structuring in our Claude training and in process automation projects.