Claude Model Pricing: Complete Token Cost Breakdown (2026)

By Abdullah Zulfiqar, Founder & Technical Editor | Last updated: October 18, 2026

Abdullah Zulfiqar is the founder of Vibe Coder Journal and an AI infrastructure benchmark researcher who tracks enterprise token economics, frontier model capabilities, and LLM developer tooling.

Quick Answer: How Much Does Claude API Cost?

Claude API pricing starts at $0.10 per million tokens (MTok) for input and $0.50 per MTok for output on Claude Haiku 5.5, scales to $2.00 / $10.00 per MTok on the flagship workhorse Claude Sonnet 5.5, and reaches $4.00 / $20.00 per MTok on Claude Opus 5.5. High-stakes reasoning tiers like Claude Fable 5.1 and Claude Mythos 5.1 are priced at $10.00 / $50.00 per MTok. All active Claude 5.5 models support a massive 1,000,000 token context window (with zero long-context surcharge on Sonnet and Opus), while prompt caching reduces cache read costs by 90% to 95%, and the Batch API provides an across-the-board 50% discount.

Whether you are architecting autonomous coding agents with vibe coding platforms, running automated evaluations, or deploying production customer support chatbots, selecting the right model and utilizing prompt caching can reduce your operational bills by over 85%.

Claude Model Pricing Table (Active Models)

The table below details Anthropic’s current active API lineup, sorted from newest frontier releases to active legacy models. All prices are in USD per million tokens (MTok). Every figure is verified against Anthropic’s official API pricing documentation.

*Note: Claude Mythos 5.1 requires organization verification via Anthropic’s specialized security programs (such as the Cyber Verification Program). On the Message Batches API, Claude Opus 5.5 and Sonnet 5.5 support up to 300,000 output tokens using the output-300k-2026-03-24 beta header.

How Prompt Caching and the Batch API Save Up to 95%

Two key features fundamentally alter Claude API economics: Prompt Caching and the Message Batches API. Understanding how their math works is essential to avoiding massive overspending.

1. Prompt Caching Economics

Prompt caching allows developers to checkpoint static or repetitive portions of their prompts—such as large system prompts, codebase indexes, documentation libraries, or few-shot examples. Instead of paying the full input token price on every API call, you write the prefix once into cache and subsequently pay only a tiny cache read fee.

• 5-Minute vs 1-Hour Cache Writes: Writing to the 5-minute cache costs 1.25x the base input rate ($2.50/M on Sonnet 5.5). Writing to the 1-hour cache costs 2.0x the base input rate ($4.00/M on Sonnet 5.5). Subsequent cache hits automatically refresh the cache TTL.
• 0.05x Cache Read Multiplier (95% Discount): On Claude Opus 5.5 and Claude Sonnet 5.5, cache hits cost only 5% of the base input price. On Sonnet 5.5, reading cached tokens drops from $2.00/M to just $0.10/M. On Opus 5.5, it drops from $4.00/M to $0.20/M.
• 0.025x Cache Read Multiplier (97.5% Discount): On Claude Fable 5.1 and Mythos 5.1, cache hits cost just 2.5% of base input ($0.25/M).
• 0.10x Cache Read Multiplier (90% Discount): On Claude Haiku 5.5, cache hits cost 10% of base input ($0.01/M for standard prompts under 100K tokens).

2. Message Batches API (Flat 50% Off)

For background tasks that do not require real-time human interaction—such as nightly batch testing, large repository re-indexing, offline synthetic data generation, or bulk document categorization—Anthropic’s Batch API provides an unconditional 50% discount across both input and output tokens. Responses are processed asynchronously within 24 hours.

The Long-Context Pricing Tier Explained

One of the most critical structural nuances in Anthropic’s pricing model is the distinction between how Haiku and Sonnet / Opus handle long context windows:

• Claude Haiku 5.5 Surcharge: Haiku 5.5 enforces a sharp pricing step function at 100,000 tokens. Prompts containing up to 100,000 input tokens cost $0.10/M input and $0.50/M output. Once your prompt exceeds 100,000 tokens (even by 1 token), the entire request jumps to the long-context tier: $0.50/M input (5x increase) and $2.50/M output (5x increase). All tokens—including cache hits and tool definitions—count toward this 100k threshold.
• Zero Surcharge on Opus 5.5 and Sonnet 5.5: Unlike Haiku, Claude Opus 5.5 and Claude Sonnet 5.5 feature a completely flat pricing curve across their full 1,000,000 token context window. A massive 900,000 token prompt on Sonnet 5.5 is billed at the exact same $2.00/M input rate as a concise 500 token query.

Three Real-World Cost Examples (With Full Math)

To see how token pricing translates into real monthly expenditures, here are three concrete scenarios with transparent step-by-step arithmetic.

Example 1: Low-Volume Web Automation Agent (Claude Sonnet 5.5)

A solo developer runs an automated browser agent performing 50 scheduled web extraction jobs per day. Each job sends 4,000 input tokens (system instructions + scraped HTML) and generates 1,000 output tokens of structured JSON.

• Daily Input Volume: 50 requests × 4,000 tokens = 200,000 tokens (0.2 MTok).
• Input Cost: 0.2 MTok × $2.00/MTok = $0.40 / day.
• Daily Output Volume: 50 requests × 1,000 tokens = 50,000 tokens (0.05 MTok).
• Output Cost: 0.05 MTok × $10.00/MTok = $0.50 / day.
• Total Spend: $0.40 + $0.50 = $0.90 per day ($27.00 per month).

Example 2: Full Codebase Refactor via Batch API (Claude Opus 5.5)

An engineering team wants Claude Opus 5.5 to audit and refactor 10 repository services overnight. Each repository context (code files, schemas, and test suites) totals 250,000 input tokens, producing 20,000 tokens of refactored code and architectural analysis per service.

• Total Tokens: 10 repos × 250,000 = 2,500,000 input tokens (2.5 MTok); 10 repos × 20,000 = 200,000 output tokens (0.2 MTok).
• Standard Real-Time Cost: (2.5 × $4.00) + (0.2 × $20.00) = $10.00 + $4.00 = $14.00.
• With Batch API (50% Off): (2.5 × $2.00) + (0.2 × $10.00) = $5.00 + $2.00 = $7.00 total.
• Savings: 50% discount ($7.00 saved) for submitting the job as an overnight batch.

Example 3: Enterprise Support Chatbot with Prompt Caching (Claude Sonnet 5.5)

A SaaS company runs a customer support chatbot answering 10,000 user inquiries per day. The chatbot uses a comprehensive 20,000-token system prompt containing product manuals and troubleshooting policies. Each user inquiry adds 500 new tokens and yields a 300-token response.

Scenario A: Without Prompt Caching

• Daily Input: 10,000 × 20,500 tokens = 205,000,000 tokens (205 MTok) × $2.00/MTok = $410.00/day.
• Daily Output: 10,000 × 300 tokens = 3,000,000 tokens (3 MTok) × $10.00/MTok = $30.00/day.
• Total Daily Bill: $410.00 + $30.00 = $440.00 per day ($13,200.00 per month).

Scenario B: With 1-Hour Prompt Caching (Refreshed Hourly across 24 Hours)

• Cache Writes: 24 hourly writes × 20,000 tokens = 480,000 tokens (0.48 MTok) × $4.00/MTok = $1.92/day.
• Cache Hits (Reads): 10,000 queries × 20,000 tokens = 200,000,000 tokens (200 MTok) × $0.10/MTok (0.05x multiplier) = $20.00/day.
• Uncached New Input: 10,000 queries × 500 tokens = 5,000,000 tokens (5 MTok) × $2.00/MTok = $10.00/day.
• Daily Output: 3 MTok × $10.00/MTok = $30.00/day.
• Total Daily Bill: $1.92 + $20.00 + $10.00 + $30.00 = $61.92 per day ($1,857.60 per month).
• Net Financial Impact: A staggering 86% cost reduction, saving $11,342.40 each month.

Legacy and Deprecated Claude Models

If you maintain legacy codebases or track multi-year LLM contracts, the table below lists older Claude models, their API availability status, and their historical rates.

Claude API vs. Claude Subscriptions (Free, Pro, Team, Enterprise)

Developers frequently wonder whether they should pay for a Claude Pro web subscription or fund an Anthropic API account. While both utilize the same underlying intelligence, their billing, feature sets, and limitations differ significantly.

If you are using tools like Claude Code, Cursor, or custom IDE extensions, check out our guide on how to see Claude Code usage to monitor your terminal token spend.

Competitor Comparison: Claude vs. GPT-6, Gemini, DeepSeek, and Grok

How does Claude model pricing compare against the rest of the frontier AI landscape? Here is how Anthropic’s lineup stacks up against OpenAI, Google, DeepSeek, and xAI across equivalent intelligence tiers.

For independent evaluation scores and SWE-bench coding leaderboards across these exact models, explore our comprehensive LLM benchmark news tracker.

Calculate Your Exact API Costs

💡 Estimate Your Monthly Spend: Want to simulate your workload with custom token lengths, caching hit ratios, and batch discounts? Use our free interactive LLM Token Cost Calculator to model exact costs across Claude, OpenAI, Gemini, and DeepSeek in real time.

Frequently Asked Questions About Claude Model Pricing

How does Claude API billing work?

Anthropic bills the Claude API on a pay-as-you-go per-token basis. You add billing credits or a payment card to your Anthropic Console account. Every API call reports usage.input_tokens, usage.output_tokens, and cache metrics in the response payload. You only pay for what you process, with zero monthly minimum platform fees.

What is the difference between Claude Sonnet 5.5 and Opus 5.5 pricing?

Claude Sonnet 5.5 costs $2.00/M input and $10.00/M output, making it the most cost-effective option for 90% of development workloads, including coding agents and complex document workflows. Claude Opus 5.5 costs $4.00/M input and $20.00/M output (exactly 2x Sonnet), reserved for deep architectural reasoning, open-ended research synthesis, and complex multi-agent orchestration.

How does prompt caching reduce Claude API costs?

Prompt caching saves static prefixes (like system prompts or large documentation files) in memory. On Claude Opus 5.5 and Sonnet 5.5, reading from cache costs only 5% of base input ($0.20/M and $0.10/M respectively). For chat systems and automated agents with recurring prompts, caching typically slashes overall token bills by 80% to 90%.

Is there a free tier for the Claude API?

No, Anthropic does not provide an ongoing free API tier. New accounts frequently receive $5 in one-time credits to test endpoints. For continuous free AI assistance, users typically rely on the free tier of the Claude web chat or open-weights models.

What is the long-context pricing tier on Claude Haiku 5.5?

While Claude Sonnet 5.5 and Opus 5.5 offer flat pricing up to 1M tokens, Claude Haiku 5.5 implements a threshold at 100,000 tokens. Prompts with up to 100,000 tokens cost $0.10 input / $0.50 output per million. Prompts exceeding 100,000 tokens are billed at $0.50 input / $2.50 output per million.

How much can I save using Anthropic’s Batch API?

The Message Batches API grants an unconditional 50% discount on both input and output tokens across every Claude model. Requests are processed asynchronously within 24 hours, making it ideal for evaluations, offline re-indexing, and non-blocking background data jobs.

Should I pay for Claude Pro or use the Claude API?

Choose Claude Pro ($20/month) if you want a seamless chat interface with Artifacts and Projects for personal productivity. Choose the Claude API if you want to integrate Claude into IDEs (like Cursor), run terminal agents (like Claude Code), or build programmatic software applications.

Sources and Verification Dates

In accordance with Vibe Coder Journal’s factual verification standard, every price, context limit, and discount structure in this report was verified on October 18, 2026 from official laboratory documentation:

• Anthropic Official Claude API Pricing Documentation (Verified: October 18, 2026)
• Anthropic Models Overview & Specifications (Verified: October 18, 2026)
• OpenAI API Pricing (Verified: October 18, 2026)
• Google Gemini API Pricing (Verified: October 18, 2026)
• DeepSeek API Official Pricing (Verified: October 18, 2026)
• xAI API Official Pricing (Verified: October 18, 2026)