How Claude API pricing works
Every request is billed on the tokens it reads and the tokens it writes. A Claude request costs input tokens times the input price, plus output tokens times the output price, with cached input billed at a discount. Five prices decide the bill:
- Input: everything you send, every time you send it: system prompt, tool definitions, chat history, retrieved documents and tool results.
- Output: everything the model writes, including extended thinking. Output costs 5× input on every current Claude model.
- Cache writes: 1.25× input to write a prompt prefix to the 5-minute cache, 2× for the 1-hour cache.
- Cache reads: 0.1× input on most models, 0.05× on Opus 5.5 and Sonnet 5.5, 0.025× on Fable 5.1.
- Batch: 50% off input and output for requests that can wait up to 24 hours. It stacks with prompt caching.
In chat and agent apps, input is usually most of the bill: each call re-sends the whole conversation and every tool result so far, while the answer stays a few hundred tokens. Count a real prompt with the token counter before estimating.
What each Claude model costs for the same job
The same workload costs up to 100× more on the largest model than on the smallest. Take 1 million classification requests a month, each with 1,000 input tokens and a 20-token answer:
| Model | Input cost | Output cost | Monthly total | Typical work |
|---|---|---|---|---|
| Claude Fable 5.1 | $10,000 | $1,000 | $11,000 | Frontier reasoning, long-horizon planning |
| Claude Opus 5.5 | $4,000 | $400 | $4,400 | Hard coding and agent tasks |
| Claude Sonnet 5.5 | $2,000 | $200 | $2,200 | Most RAG answers, coding, agent steps |
| Claude Haiku 5.5 | $100 | $10 | $110 | Classification, extraction, routing |
Route each request type to the smallest model that passes your evaluation set. The LLM cost optimization guide shows routing, caching and context cuts stacked on one monthly bill.
Prompt caching and batch prices
Prompt caching bills a repeated prompt prefix at a fraction of the input price after the first write. On Sonnet 5.5 a cached read is $0.10 per million tokens against $2 uncached, 95% off. A 5-minute write costs 1.25× input once, so a prefix pays for itself as soon as one later request reads it. The Anthropic prompt caching guide covers the break-even math, minimum lengths and what breaks a cache.
The Batch API halves input and output prices for asynchronous work: Sonnet 5.5 drops to $1 input and $5 output. Caching discounts apply on top, so a cached read in a batch costs half the cached price.
Other charges on the Claude API
- Long context: Claude 4.6 and later models bill their full 1M-token context window at standard prices, except Haiku 5.5, whose prompts over 100K tokens pay $0.50 input and $2.50 output.
- Fast mode: faster output on Opus 5.5 at $8 input and $40 output per million tokens. Not available with the Batch API.
- US-only inference: setting
inference_geoto US multiplies every token price by 1.1 on Claude 4.6 and later. - Web search: $10 per 1,000 searches, plus the tokens of the results, which stay in the context for later turns.
- Tokenizer: Claude 4.7 and later use a tokenizer that produces about 30% more tokens for the same text, so compare costs on real prompts, not on price per token alone.
Older Claude models
Still available on the API at these prices. Newer models in the same tier are usually as cheap or cheaper for better results.
| Model | Input | Output | Cached input | Cache write | Batch input | Batch output |
|---|---|---|---|---|---|---|
Claude Fable 5 | $10.00 | $50.00 | $1.00 | $12.50 | $5.00 | $25.00 |
Claude Opus 5 | $5.00 | $25.00 | $0.50 | $6.25 | $2.50 | $12.50 |
Claude Opus 4.8 | $5.00 | $25.00 | $0.50 | $6.25 | $2.50 | $12.50 |
Claude Opus 4.7 | $5.00 | $25.00 | $0.50 | $6.25 | $2.50 | $12.50 |
Claude Opus 4.6 | $5.00 | $25.00 | $0.50 | $6.25 | $2.50 | $12.50 |
Claude Opus 4.5 | $5.00 | $25.00 | $0.50 | $6.25 | $2.50 | $12.50 |
Claude Sonnet 5 | $2.00 | $10.00 | $0.20 | $2.50 | $1.00 | $5.00 |
Claude Sonnet 4.6 | $3.00 | $15.00 | $0.30 | $3.75 | $1.50 | $7.50 |
Claude Sonnet 4.5 | $3.00 | $15.00 | $0.30 | $3.75 | $1.50 | $7.50 |
Claude Haiku 4.5 | $1.00 | $5.00 | $0.10 | $1.25 | $0.50 | $2.50 |
Frequently asked questions
How much does the Claude API cost?
Claude API prices run from $0.10 per million input tokens and $0.50 per million output tokens on Claude Haiku 5.5 to $10 and $50 on Claude Fable 5.1. Sonnet 5.5 costs $2 and $10, and Opus 5.5 costs $4 and $20. Cached input and the Batch API reduce these prices.
How much does one token cost on Claude?
Prices are quoted per million tokens, so divide by 1,000,000. On Claude Sonnet 5.5 one input token costs $0.000002 and one output token $0.00001, so a request with 1,000 input and 500 output tokens costs $0.007. On Claude Haiku 5.5 the same request costs $0.00035, on Claude Fable 5.1 $0.035.
How much is Claude Sonnet 5.5 per million tokens?
$2 per million input tokens and $10 per million output tokens. Cached reads cost $0.10, a 5-minute cache write $2.50, and the Batch API halves input and output to $1 and $5.
How much is Claude Opus 5.5 per million tokens?
$4 per million input tokens and $20 per million output tokens, with cached reads at $0.20. Fast mode costs $8 input and $40 output. The Batch API brings standard prices down to $2 and $10.
Is the Claude API free?
No. The API is pay-as-you-go per token; Anthropic gives new accounts a small amount of free credits to test it. Claude Haiku 5.5 is the cheapest model at $0.10 per million input tokens.
Is the Claude API included in a Claude Pro or Max subscription?
No. Claude subscriptions cover the Claude apps. The API is billed separately, per token, through the Claude Console.
Do Claude's batch and caching discounts stack?
Yes. Batch processing halves input and output prices, and prompt caching multipliers apply on top, so a cached read in a batch request costs half the normal cached price.
Why does the same prompt use more tokens on newer Claude models?
Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. A price per token is only comparable after counting the prompt on that model.