Anthropic Claude API Pricing 2026: Token Rates, Plans, and Full Cost Breakdown

If you are looking for Anthropic Claude API pricing in 2026, this guide covers everything: exact token rates for all three Claude model families, the 2026 price cuts that changed the value proposition, prompt caching savings of up to 90%, Batch API discounts, and every consumer subscription plan from Free to Enterprise.

Whether you are comparing Claude AI pricing for a subscription or evaluating the Claude API for a production application, the numbers below are current as of May 2026.

---

Claude Pricing at a Glance

---

Anthropic Claude API Pricing 2026: Exact Token Rates

The Anthropic API (formerly Claude API) gives developers pay-per-token access to all Claude models. Pricing is per million tokens, with input and output billed at different rates.

Claude API Rates (Per Million Tokens)

What Each Claude API Model Is For

Claude Haiku 4.5 at $1.00/$5.00 per million tokens is Claude's fastest and most cost-efficient model. Best suited for high-volume tasks where reasoning depth is less critical: summarisation, classification, structured data extraction, customer support routing, and simple Q&A. At $1.00/M input, running 1,000 short customer support interactions costs approximately $2–$10 total.

Claude Sonnet 4.6 at $3.00/$15.00 is the optimal balance model for most production applications. It handles complex writing, multi-step reasoning, code generation, and detailed analysis — tasks where Haiku falls short — at a lower cost than Opus. Most teams building AI products default to Sonnet 4.6 as their primary model.

Claude Opus 4.7 at $5.00/$25.00 is Anthropic's most capable model and the right choice for genuinely hard problems: complex multi-step reasoning, advanced code generation, nuanced analysis with many constraints, and tasks where maximum intelligence is the bottleneck rather than throughput or cost. Following the February 2026 price reduction, Opus 4.7 is now accessible for production use cases where it was previously too expensive.

Practical Cost Examples

• Processing a 10-page document (5,000 tokens) with Sonnet 4.6: $0.015
• Generating a 500-word article (700 output tokens) with Opus 4.7: $0.0175
• Running 1,000 customer support responses with Haiku 4.5 (2M total tokens): $2–$10
• Processing a 100,000-token codebase with Sonnet 4.6: $0.30 input cost

---

Claude API Prompt Caching: Up to 90% Savings

Prompt caching is the most impactful cost reduction available on the Anthropic Claude API. If your application sends the same large system prompt, document, or knowledge base on every request, you pay the full input rate only once — subsequent uses of the cached content cost 10% of the input rate.

How prompt caching works:

• First request: full input token rate + cache write surcharge
• Subsequent requests using cached content: cache read rate (10% of input price)
• Cache TTL: 5 minutes, reset with each use

Cache read rates vs full input rates:

Real-world example: A customer service application with a 10,000-token system prompt running 500 requests per day:

• Without caching: 10,000 × 500 × $3.00/M = $15.00/day in system prompt tokens alone
• With caching: $3.75 (one-time write) + 499 × $0.30/M × 10,000 = $1.50/day
• Result: 90% reduction, saving $13.50/day or ~$405/month

For any API application with a repeated large context, prompt caching is non-negotiable.

---

Claude Batch API: 50% Discount on Non-Urgent Requests

The Anthropic Batch API processes requests asynchronously and returns results within 24 hours. In exchange for the wait, you receive a 50% discount on both input and output tokens.

Batch API Rates (Per Million Tokens)

Best use cases for Batch API:

• Processing large document libraries or archives
• Nightly data analysis pipelines
• Generating product descriptions or metadata at scale
• Running evaluation datasets for AI research and testing
• Any workload where results are not needed in real time

At $1.50/$7.50 for Sonnet 4.6 via Batch API, it becomes cheaper than GPT-4o ($2.50/$10.00) at list price for equivalent-tier workloads.

---

Claude Subscription Plans: Full Breakdown

Claude Free

Claude's free tier gives access to Claude Sonnet with rolling usage limits. No credit card required, no trial expiry — free indefinitely.

Free plan includes:

• Claude Sonnet access (not Opus 4.7)
• Rolling daily usage limits (significantly lower than paid plans)
• Basic Projects functionality
• No Claude Code
• No extended thinking
• No priority access during peak hours

The free plan covers occasional use, short conversations, and evaluating Claude before subscribing. For regular daily professional use, limits become a bottleneck.

Claude Pro — $20/Month

Pro is the right plan for most individual power users and professionals.

Claude Pro includes, over free:

• 5x the usage volume of the free tier (rolling)
• Access to all Claude models including Claude Opus 4.7
• Claude Code — file creation, code execution, and full agentic task capability
• Unlimited Projects
• Google Workspace integration
• Remote MCP (Model Context Protocol) connectors
• Extended thinking mode
• Priority access during high-traffic periods
• Voice mode

At $20/month, Claude Pro matches ChatGPT Plus ($20/month) in price and includes Claude Code — a significant capability advantage for developers that ChatGPT Plus does not directly replicate.

Limitation: Anthropic does not publish exact message counts. Limits vary by model — heavy Opus 4.7 usage depletes your allowance faster than Sonnet or Haiku.

Claude Max — $100/Month and $200/Month

Max plans are for users who regularly hit Pro limits.

Max 5x — $100/month:

• 5x the usage of Pro (25x free tier)
• All Pro features
• Same models, same features — more headroom

Max 20x — $200/month:

• 20x the usage of Pro (100x free tier)
• All Pro and Max 5x features
• Designed for users running Claude as their primary work tool for multiple hours daily, or developers running extended Claude Code agentic sessions

Both Max plans support 100K+ token context windows. As of March 2026, the 1M token context window is included at no surcharge for Max plan users — previously a premium add-on.

Claude Team Plans

Team plans add collaboration, admin controls, and higher per-seat limits.

Team Standard — $25/seat/month (minimum 5 seats):

• All Pro features per seat
• Admin dashboard and usage controls
• SSO (Single Sign-On) support
• Domain capture for automatic team onboarding
• Slack, Microsoft 365, and Google Workspace integrations
• Centralised billing
• Does not include Claude Code

Team Premium — $100/seat/month (minimum 5 seats):

• All Team Standard features
• Full Claude Code access per seat
• 6.25x Max 5x headroom per session
• Priority support

Minimum 5-seat requirement: Team Standard costs $125/month minimum; Team Premium costs $500/month minimum.

Claude Enterprise

Enterprise is custom-priced via direct contract with Anthropic.

Enterprise adds over Team:

• 500K token context window (no surcharge as of March 2026)
• 1M token context window available
• HIPAA readiness for healthcare organisations
• Custom compliance and security controls
• Dedicated account management
• Custom rate limits negotiated per organisation
• Advanced audit logging and data residency options

Enterprise targets healthcare companies (HIPAA), financial services (data residency), and large organisations running Claude at significant API volume with contractual data handling requirements.

---

The 2026 Claude Price Cuts

Two changes in early 2026 substantially improved Anthropic Claude API pricing:

Opus 4.7 price reduction (67% cut, February 2026): Opus 4.7 was previously priced at $15 input / $75 output per million tokens. The February 2026 cut to $5/$25 made frontier-model quality economically viable for production applications. At the old price, Opus usage at any real volume was cost-prohibitive for most companies.

1M token context window at no surcharge (March 2026): The 1M token context window was previously a premium add-on. As of March 2026, it is included at no extra cost for Enterprise and Max plan users — making Claude significantly more capable for processing entire codebases, legal document libraries, or research corpora in a single prompt.

---

Claude API vs GPT-4o vs Grok API: Price Comparison

Low-Cost Models

Claude Haiku 4.5 is more expensive than GPT-4o Mini and Grok's fast-tier models at list price, but is significantly more capable for tasks requiring nuanced language understanding.

Frontier Models

At list prices, Claude Sonnet 4.6 and Opus 4.7 are more expensive than GPT-4o and Grok 4.3. However:

• Prompt caching on Sonnet 4.6 brings the effective input cost to $0.30/M for repeated context — cheaper than GPT-4o at $2.50/M
• Batch API on Sonnet 4.6 at $1.50/$7.50 is cheaper than GPT-4o at list price
• Claude's benchmark performance on reasoning, writing, and coding tasks frequently exceeds GPT-4o, making the cost-per-quality comparison more favourable than raw token prices suggest

The best model for a given application depends on the task, not just the price.

---

Which Claude Plan Should You Choose?

Choose Claude Free if:

• You use AI a few times per week for non-critical tasks
• You are evaluating Claude before committing to a subscription
• Your tasks are short and do not require Opus-level reasoning

Choose Claude Pro ($20/month) if:

• You use Claude daily for professional work
• You need Claude Code for software development or agentic tasks
• You need Opus 4.7 access for complex reasoning
• You regularly work with long documents or large codebases

Choose Claude Max 5x ($100/month) if:

• You regularly hit Pro usage limits mid-day
• You run long Claude Code agentic sessions
• You need 100K+ context windows consistently

Choose Claude Max 20x ($200/month) if:

• Claude is your primary work tool for 6+ hours daily
• You run complex multi-step agentic tasks every day
• You are a developer or researcher at the frontier of AI-assisted work

Choose Claude Team ($25–$100/seat/month) if:

• You are buying for a team of 5+ and need admin controls and SSO
• You need centralised billing and usage visibility
• Your team uses Claude as a core work tool

Choose the Claude API if:

• You are building an application that uses Claude programmatically
• You need per-token control over model selection and cost
• Your usage is variable and token pricing beats a flat subscription

---

Tips to Reduce Claude API Costs

Match model to task. Haiku 4.5 handles summarisation, classification, and simple Q&A. Sonnet 4.6 covers complex writing, reasoning, and code. Reserve Opus 4.7 for genuinely hard problems. Defaulting to Opus on every request is the most common and expensive mistake.

Implement prompt caching. Any application with a repeated system prompt or knowledge base should use prompt caching — the 90% savings on cached tokens is the highest-ROI optimisation available.

Use Batch API for overnight workloads. The 50% discount compounds significantly at scale. Any pipeline that does not need real-time results should go through the Batch API.

Write concise system prompts. Every token in your system prompt is billed on every non-cached request. Cutting a 2,000-token system prompt to 500 tokens reduces that cost by 75% at scale.

---

Frequently Asked Questions

What is Claude API pricing in 2026? Claude Haiku 4.5 costs $1.00/$5.00 per million input/output tokens. Claude Sonnet 4.6 costs $3.00/$15.00. Claude Opus 4.7 costs $5.00/$25.00. Prompt caching reduces effective input costs by up to 90%.

How much does Claude Pro cost? Claude Pro costs $20/month. It includes access to Claude Opus 4.7, Claude Code, extended thinking, and 5x the message volume of the free tier.

Is Claude API cheaper than OpenAI? At list prices, Claude Sonnet 4.6 ($3.00/$15.00) is more expensive than GPT-4o ($2.50/$10.00). With prompt caching, the effective Sonnet 4.6 cost for repeated-context applications drops to $0.30/M input — below GPT-4o list price. Grok 4.3 ($1.25/$2.50) is cheaper than both at list price.

What is prompt caching and how much does it save? Prompt caching lets you reuse a cached system prompt or document across multiple API calls. Subsequent uses cost 10% of the standard input rate — a 90% saving on those tokens.

What is the Batch API discount? The Anthropic Batch API processes requests asynchronously (results within 24 hours) at 50% off both input and output token rates. Sonnet 4.6 drops from $3.00/$15.00 to $1.50/$7.50.

What context window does Claude support? Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 all support 100K+ token context windows via API. Enterprise and Max plan users have access to 500K and 1M token context windows at no surcharge as of March 2026.

How does Claude compare to Grok API for developers? Grok 4.3 ($1.25/$2.50) has lower list prices than Claude Sonnet 4.6 ($3.00/$15.00). Claude's advantage is prompt caching (reduces effective cost to $0.30/M input for repeated context), stronger benchmark performance on complex reasoning tasks, and Claude Code for agentic development workflows.

---

The Bottom Line

Anthropic Claude API pricing in 2026 is tiered to match different workloads. The February 2026 price cut made Opus 4.7 viable for production. Prompt caching and Batch API discounts make Sonnet 4.6 cost-competitive with GPT-4o for applications with repeated context. The subscription plans — especially Claude Pro at $20/month — are competitively priced and include Claude Code, a meaningful capability advantage for developers.

The core decision: subscription for predictable monthly cost and individual or team use; API for programmatic access with per-token control. For developers, always prototype with the Batch API enabled and prompt caching implemented before comparing costs with competitors.

---

Building an AI startup? List it on Startup Launch Page and reach investors and early adopters who are actively looking for what you're building.