Claude Model Tokens

Claude 3.5 Sonnet Context Window Tokens

Reference the 200K token context window, estimate prompt fit, and calculate Claude 3.5 Sonnet API token cost.

Last updated: June 17, 2026

Feature Comparison

ItemValuePlanning note
Context window200,000 tokensInput plus working context must fit before the response.
Input price$3 per 1M tokensUse for prompts, files, messages, and tool context.
Output price$15 per 1M tokensUse for generated response tokens.
Safe planning reserve10% to 20%Leave room for formatting, tools, and completion.
Best fit checkOfficial token counting APIUse count_tokens for structured prompts before sending.

Context Window Calculator

Use this formula when planning a long prompt, repo summary, document batch, or tool-heavy request.

usable_context = 200000 - expected_output_tokens - safety_reserve
remaining_input = usable_context - system_tokens - message_tokens - tool_tokens

Cost Calculator

For Claude 3.5 Sonnet API pricing, estimate total cost with separate input and output token rates.

input_cost = input_tokens / 1000000 * 3
output_cost = output_tokens / 1000000 * 15
total_cost = input_cost + output_cost

Example:
input_tokens = 80000
output_tokens = 4000
total_cost = 80000 / 1000000 * 3 + 4000 / 1000000 * 15
           = $0.30

Practical Planning Table

These rough buckets help decide whether to paste content, summarize it, or use retrieval.

  • Under 20K tokens: usually safe for direct prompts and a full response.
  • 20K to 100K tokens: leave deliberate output room and avoid unnecessary logs.
  • 100K to 180K tokens: plan reserves carefully and count tokens before sending.
  • Above 180K tokens: summarize, chunk, retrieve, or reduce tool output first.

Official References

Anthropic's Claude 3.5 Sonnet launch post lists the 200K context window and launch pricing. The token counting docs explain how to count structured message tokens before sending.

FAQ

What is the Claude 3.5 Sonnet context window?

Claude 3.5 Sonnet launched with a 200,000 token context window.

Does output count against the context window?

You should reserve context for the expected response. If the input fills the window, the model has too little room to answer.

How much does Claude 3.5 Sonnet cost per token?

The launch pricing was $3 per million input tokens and $15 per million output tokens.

How do I count Claude tokens accurately?

Use Anthropic's token counting endpoint for structured API messages. Character or word-count heuristics are only rough estimates.