Claude pricing per million tokens
| Model | Input | Output | 5m write | 1h write | Cache read |
|---|---|---|---|---|---|
| Claude Sonnet 5.5 | $2 | $10 | $2.5 | $4 | $0.2 |
| Claude Opus 5.5 | $4 | $20 | $5 | $8 | $0.2 |
| Claude Haiku 4.5 | $1 | $5 | $1.25 | $2 | $0.1 |
Rates are a checked snapshot for standard, direct Claude API use in USD. This calculator supports the three listed models. It excludes batch discounts, fast mode, US-only inference premiums, server-tool fees, taxes and private contracts. Other platforms may charge different rates. Consult the official price source for those cases.
How the calculation works
Enter each category separately. Uncached input excludes the tokens entered as cache reads or writes. Output is a quantity you supply, not a prediction from your prompt. The five category costs sum to one request; monthly cost repeats that request the number of times you enter.
Request cost = Σ(category tokens × category rate / 1,000,000)
Monthly estimate = request cost × requests per month
An illustrative Sonnet 5.5 request with 10,000 uncached input tokens and 1,000 output tokens costs $0.03 at the listed standard rates. Repeating it 1,000 times gives a $30 planning estimate. This example assumes no caching and the same output on every request.
Plan with output and cache separately
Use the “Cached request” example to explore a repeat request with 1,000 fresh input tokens, 9,000 cache-read tokens and 500 output tokens. It assumes the prefix was cached earlier. Include the write cost in the request that creates it; cache reads and cache writes are separate quantities.
A monthly scenario is only as useful as its assumptions. When testing a real workflow, gather observed output and cache quantities from representative requests. Use the low and high cases you actually see instead of treating one short demo as the monthly average.
Can I use the local Claude count?
The homepage's Claude option is a rough legacy-tokenizer reference. Choosing a pricing model does not make that reference match a modern Claude tokenizer. Counts transferred from that option keep an estimate warning. Use the model-specific counting API or observed usage before relying on a production budget.
The calculator accepts manual quantities as well. See methodology and accuracy to understand what the local tools measure.
API cost and a subscription answer different questions
A per-token scenario helps plan API spending. It does not tell you how much Free, Pro or Max usage remains. For product limits, start with Claude token limits. For a coding session, use Claude Code usage to locate the relevant usage information.
Common questions
Is this a live price feed?
No. The source and check date identify a maintained snapshot. Recheck the provider's pricing when making a purchasing decision.
Why can a tiny request show several decimal places?
Prices are quoted per million tokens. A small request can cost much less than one cent. Extra decimal places keep those estimates visible without rounding them down to zero.
Does cached input disappear from the context window?
Caching changes its price. It still occupies context. Use the limits page when checking whether a planned request fits.
Sources & review
Checked 3 October 2026. Product rules can change; consult the linked official references.