Free tool · Anthropic list prices, checked 28 October 2026
What a product built on Claude costs to run, per output and per customer. Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens; the bill that matters is your instructions read on every run, which prompt caching cuts to 5% of the input price.
Built a method in Claude you want to sell? We measure the real cost on your own sample matters and turn it into a product. Free, under NDA.
Most calculators price a chat message. A product built from a lawyer's or consultant's method is different: the method is long, and the model reads all of it on every run, before it reads the customer's document.
Take a method of 400,000 tokens, a 50-page document and a 2,000-word draft on Claude Opus 5.5. Read in full every time, one run costs $1.78. With the method cached, the same run costs $0.262, and on Claude Sonnet 5.5 with caching, $0.131. At 20 runs a month on Opus with caching, that customer costs about $5.24 a month in model fees.
Caching is the setting that decides whether the business works. It only pays off when the instructions are byte-for-byte identical on every run and sit at the start of the prompt, so the product has to be built that way from the first version.
One run, Claude Opus 5.5
Anthropic list prices, 28 October 2026. 650 tokens a page.
The current models, from Anthropic's pricing page, read 28 October 2026. The Batch API halves every column.
| Model | Input | Cache write, 1 hour | Cache read | Output |
|---|---|---|---|---|
| Claude Opus 5.5 | $4 | $8 | $0.2 | $20 |
| Claude Sonnet 5.5 | $2 | $4 | $0.1 | $10 |
| Claude Haiku 5.5prompts up to 100K tokens | $0.1 | $0.2 | $0.01 | $0.5 |
| Claude Fable 5.1 | $10 | $20 | $0.25 | $50 |
| Claude Haiku 5.5prompts over 100K tokens | $0.5 | $1 | $0.05 | $2.5 |
How the calculator works: pages become tokens at 500 words a page and 1.3 tokens a word. With caching on, the instructions are billed as a cache read on every run, and the product pays a one-hour cache write when the cache has gone cold: about once a working day when runs come at least hourly, up to once per busy hour when they are sparse (8 busy hours a day, 22 days a month). The calculator spreads that cost across your customers. The document and the output are never cached. Thinking tokens bill as output and vary by task, so treat the output figure as a floor.
[ FAQ ]
Claude API pricing is per million tokens, separately for input and output. As of 28 October 2026: Claude Opus 5.5 is $4 input and $20 output, Claude Sonnet 5.5 $2 and $10, Claude Haiku 5.5 $0.10 and $0.50 for prompts up to 100,000 tokens, and Claude Fable 5.1 $10 and $50. Cached input is read at a fraction of the input price, and the Batch API takes 50% off input and output.
About 650 for a page of dense legal or business text: roughly 500 words at about 1.3 tokens a word. Tables, citations and non-English text run higher. For an exact figure, Anthropic's token counting endpoint counts a real document before you send it.
The part of the prompt that repeats on every run, your instructions, is written to the cache once and then read at a fraction of the input price: $0.20 instead of $4 per million tokens on Claude Opus 5.5. A one-hour cache write costs twice the input price, and the cache is shared across your customers as long as the instructions are identical, so a busy product pays for the write a few times a day, not on every run.
Not on the current models. Anthropic includes the full 1M token context window at standard pricing for Claude 4.6 and later, except Claude Haiku 5.5, which charges a higher rate when the whole prompt, cached tokens included, is over 100,000 tokens. The calculator switches to that tier automatically.
For work that can wait, yes: 50% off input and output, and it stacks with caching. Results usually come back well within a day. It suits overnight reports and bulk review. It does not suit a customer watching a progress bar.
Start from the cost per customer per month, then decide the margin you need after hosting, support and the model bill. Software businesses usually aim for gross margins of 70 to 85 percent. The calculator gives the floor at the margin you choose; your market decides how far above it you can go.
