Claude API Pricing 2026

Complete Claude API pricing guide for 2026: all model prices, prompt caching savings, batch API discount, context window costs, and how to estimate your bill. With Python examples.

Claude API pricing is straightforward once you understand the three levers: model tier, prompt caching, and the Batch API. Here's everything you need to estimate and minimize your bill.

2026 Claude API pricing table

Prompt caching prices

Batch API: 50% discount for async workloads

Cost estimation in Python

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarize this 500-word article: ..."}]
)

usage = response.usage
input_cost  = usage.input_tokens  * 3.00 / 1_000_000
output_cost = usage.output_tokens * 15.00 / 1_000_000
print(f"Input:  {usage.input_tokens} tokens = ${input_cost:.6f}")
print(f"Output: {usage.output_tokens} tokens = ${output_cost:.6f}")
print(f"Total:  ${input_cost + output_cost:.6f}")

Prompt caching example — production chatbot

import anthropic

client = anthropic.Anthropic()

# 2,000-token system prompt cached across all user requests
SYSTEM = "You are a senior software engineer with 15 years of experience..." + " expert knowledge " * 400

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system=[{
        "type": "text",
        "text": SYSTEM,
        "cache_control": {"type": "ephemeral"}
    }],
    messages=[{"role": "user", "content": "How do I handle database migrations in production?"}]
)

usage = response.usage
print(f"Cache read tokens: {usage.cache_read_input_tokens}")
print(f"Cache write tokens: {usage.cache_creation_input_tokens}")

# With 2,000 cached input tokens at $0.30/M vs $3/M:
# Savings per call: 2,000 × (3.00 - 0.30) / 1,000,000 = $0.0054
# At 100K calls/month: $540/month saved

Cost comparison: models by task type

Cost optimization checklist

• Start with Haiku: Run your task on Haiku first; upgrade to Sonnet only if quality is insufficient. Haiku costs 73% less per call.
• Enable prompt caching: Any system prompt longer than ~500 tokens repeated across many calls should be cached. Add cache_control: {type: "ephemeral"} to the content block.
• Use the Batch API: For analytics, document processing, or any task that doesn't need real-time response — Batch API cuts costs 50% with no quality tradeoff.
• Set max_tokens conservatively: You're charged for actual output tokens, not max_tokens. But a low max_tokens prevents runaway long responses on edge cases.
• Log token usage: Always log response.usage in production. Unexpected token spikes are the #1 source of surprise bills.

Use the Claude API Cost Calculator for interactive estimation with your specific token volumes. For Sonnet vs GPT-4o pricing comparison, see Claude Sonnet 4.6 vs GPT-4o. For cost optimization patterns, see Claude API Cost Optimization.

Frequently asked questions

Free tools

More examples