Claude API Pricing Guide 2026: Every Model, Cost, and Saving Tip

by

About 18; min

Anthropic’s Claude API powers applications that need strong reasoning, long-context analysis, and reliable code generation. The model family spans from the lightweight Haiku to the flagship Opus, each targeting different cost-performance tradeoffs. Here’s the complete guide to Claude API pricing in 2026.

Model Pricing Overview

Model Input (per M tokens) Output (per M tokens) Context Window Best For
Claude Haiku 3.5 $0.80 $4.00 200K Fast, cheap tasks
Claude Sonnet 4 $3.00 $15.00 200K Best quality/price balance
Claude Opus 4 $15.00 $75.00 200K Maximum reasoning capability

Claude Haiku 3.5 — The Speed Tier


Claude Haiku 3.5 is Anthropic’s fastest and cheapest model. At $0.80 per million input tokens and $4.00 per million output, it handles high-volume tasks where speed and cost matter more than peak reasoning quality. The 200K context window matches larger models. Haiku excels at classification, extraction, summarization, simple Q&A, and routing tasks. Response latency is the lowest in the Claude family — ideal for real-time applications. Many production systems use Haiku as the default model, routing only complex queries to Sonnet or Opus. A customer support bot processing 50,000 conversations per month at 800 tokens each costs approximately $32 with Haiku versus $180 with Sonnet.

Claude Sonnet 4 — The Default Choice


Claude Sonnet 4 is Anthropic’s recommended model for most applications. At $3.00 per million input tokens and $15.00 per million output, it delivers near-Opus quality at one-fifth the cost. Sonnet handles complex coding tasks, detailed analysis, nuanced writing, and multi-step reasoning reliably. The 200K token context window processes entire codebases, long documents, and extended conversations. Sonnet scores within 5-10% of Opus on most benchmarks while being significantly faster and cheaper. For developers building AI products, Sonnet is the model you should evaluate first — switch to Haiku for simpler tasks or Opus for the hardest problems only after benchmarking Sonnet on your specific use case.

Claude Opus 4 — Maximum Capability


Claude Opus 4 is Anthropic’s most capable model for complex reasoning, expert-level coding, and tasks that require the deepest understanding. At $15.00 per million input tokens and $75.00 per million output, it’s 5x more expensive than Sonnet. Use Opus when the task demands the absolute best quality — complex software architecture decisions, nuanced legal analysis, advanced mathematical proofs, or multi-step reasoning chains where accuracy is critical. The 200K context window is the same as other Claude models. For most applications, the quality gap between Sonnet and Opus doesn’t justify the 5x cost increase. Reserve Opus for your hardest problems.

Additional API Features and Pricing

Feature Pricing Details
Message Batches API 50% discount Non-urgent requests, 24-hour SLA
Prompt Caching 90% discount on cached tokens Reuse long system prompts
Extended Thinking Charged as output tokens Chain-of-thought reasoning
Tool Use No extra cost Function calling included
Vision (Image Input) Token-based (image size) No extra per-image fee
Streaming No extra cost Real-time token delivery

Cost-Saving Features

Prompt Caching

Prompt Caching reduces costs by 90% for repeated context. If your application sends the same system prompt, few-shot examples, or reference documents with every request, cached tokens cost only 10% of the standard input rate. For a RAG application that includes a 10,000-token knowledge base in every query, caching saves $2.70 per million requests with Sonnet. Enable caching by marking the static portion of your prompt — subsequent requests reuse the cached version.

Message Batches API

The Batches API processes non-time-sensitive requests at 50% discount. Submit a batch of requests, and Anthropic processes them within 24 hours. Perfect for content generation, data processing, evaluation pipelines, and any workload where real-time responses aren’t required. Sonnet batch pricing drops to $1.50/M input and $7.50/M output — approaching Haiku’s standard pricing with Sonnet-level quality.

Cost Scenarios

AI chatbot (10,000 conversations/month)

Average conversation: 500 input + 400 output tokens.

  • Haiku: (5M × $0.80/M) + (4M × $4.00/M) = $4.00 + $16.00 = $20/month
  • Sonnet: (5M × $3.00/M) + (4M × $15.00/M) = $15.00 + $60.00 = $75/month
  • Opus: (5M × $15.00/M) + (4M × $75.00/M) = $75.00 + $300.00 = $375/month

Code generation tool (5,000 requests/month)

Average request: 1,000 input + 1,500 output tokens.

  • Sonnet: (5M × $3.00/M) + (7.5M × $15.00/M) = $15.00 + $112.50 = $127.50/month
  • With Batches (50% off): $63.75/month
  • With Prompt Caching (800 cached tokens/request): Saves ~$10.80/month

Document analysis (1,000 PDFs/month, avg 50 pages each)

Average: 30,000 input tokens + 2,000 output per document.

  • Sonnet: (30M × $3.00/M) + (2M × $15.00/M) = $90.00 + $30.00 = $120/month
  • With Prompt Caching (shared instructions): ~$95/month

Claude vs OpenAI API Pricing

Tier Claude OpenAI Price Difference
Budget Haiku: $0.80/$4.00 GPT-4o mini: $0.15/$0.60 OpenAI 5x cheaper
Standard Sonnet: $3.00/$15.00 GPT-4o: $2.50/$10.00 OpenAI 20-33% cheaper
Premium Opus: $15.00/$75.00 o1: $15.00/$60.00 Similar input, OpenAI cheaper output

OpenAI’s models are generally cheaper per token, especially at the budget tier (GPT-4o mini vs Haiku). Claude’s advantage is in task quality — many developers find Sonnet produces better code, more accurate analysis, and more nuanced writing than GPT-4o, reducing downstream costs from rework and debugging. The choice should be based on output quality for your specific task, not just per-token price.

Best Practices

  • Start with Sonnet — Evaluate Sonnet first. Only use Haiku if Sonnet is overkill, or Opus if Sonnet isn’t good enough.
  • Enable Prompt Caching — Any repeated context (system prompts, examples, documents) should be cached for 90% savings.
  • Use Batches for non-urgent work — Content generation, evaluation, and data processing at 50% off.
  • Set max_tokens — Prevent unnecessarily long outputs that waste expensive output tokens.
  • Monitor with usage dashboard — Track spending per model and per use case in the Anthropic Console.
Our Verdict


Claude Sonnet 4 is the right model for most API applications. It delivers near-Opus quality at one-fifth the price and handles everything from coding to analysis to writing. Claude Haiku 3.5 earns runner-up for high-volume, cost-sensitive tasks where speed matters more than peak capability. The combination of Prompt Caching and Batches API can reduce Sonnet costs by 50-90% for the right workloads, making it competitive with budget-tier models on price while maintaining premium quality.

Get Claude API Key

FAQ

Is there a free tier?

New accounts receive $5 in free API credits. After credits expire, usage is billed per token. The free Claude.ai consumer product has daily usage limits but doesn’t provide API access.

What’s the difference between Claude.ai and the API?

Claude.ai is the consumer chat product ($0-30/mo subscription). The API is for developers building applications — billed per token with no monthly subscription. Different products, different pricing.

How does Prompt Caching work?

Mark the static portion of your prompt (system instructions, examples, reference text) as cacheable. The first request pays full price. Subsequent requests reuse the cached prompt at 90% discount. Cache lasts 5 minutes by default with automatic refresh on use.

Should I use Claude or OpenAI API?

Benchmark both on your actual tasks. Claude tends to produce better code and more careful analysis. OpenAI tends to be cheaper per token, especially at the budget tier. Many production systems use both — routing tasks to whichever model performs best for each specific use case.