Claude API Cost per Million Tokens (2026)

The Claude API is billed per token, in and out, with no monthly fee. At list prices checked on 28 October 2026, Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens; Claude Haiku 5.5 costs $0.10 and $0.50, Claude Opus 5.5 $4 and $20, and Claude Fable 5.1 $10 and $50. A typical chat message of 3,000 input and 600 output tokens therefore costs about 1.2 cents on Sonnet 5.5, and 30 such messages a day come to about $11 a month: less than a $20 Claude Pro plan.

Below: the price table, what a token is, three worked examples, the plan-or-API question and the three things that move the bill most. Every number comes from the same price file our calculators use (/data/prices.json).

Claude API prices per million tokens

Model	Input $/M	Output $/M	Cache read $/M	5-min cache write $/M	Batch input / output $/M
Claude Fable 5.1	$10	$50	$0.25	$12.50	$5 / $25
Claude Opus 5.5	$4	$20	$0.20	$5	$2 / $10
Claude Sonnet 5.5	$2	$10	$0.10	$2.50	$1 / $5
Claude Haiku 5.5	$0.10	$0.50	$0.01	$0.125	$0.05 / $0.25

Checked 28 October 2026 on Anthropic’s pricing page. Prices are in USD, before tax, for the standard API tier. Two footnotes matter:

Haiku 5.5 has a second rate card. A request whose prompt is over 100,000 tokens pays $0.50 in and $2.50 out instead. Each request is priced on its own.
The Batch API halves both prices for jobs that can wait (results arrive within 24 hours). A nightly summariser or a backfill belongs there.

A cache read is an input token that was already in a recent request and is served from a prompt cache: a tenth of the normal input price on Haiku, a twentieth on Sonnet 5.5 and Opus 5.5, a fortieth on Fable 5.1. Writing to the cache costs 25% more than normal input. The prompt caching guide works through when that pays off.

What a token is

A token is a piece of text about four characters long in English, so 1,000 tokens is roughly 750 words. A page of a contract is 500 to 700 tokens; a 20-page PDF is 10,000 to 15,000. The whole prompt counts as input: your instructions, any documents, the conversation so far and the user’s message. Only the model’s reply counts as output, and output is five times the price of input on every Claude model.

That asymmetry explains most Claude bills. Chat and document work are input-heavy, so the input price dominates; code generation and long reports are output-heavy, so the output price does.

Three worked examples

Costs below are per request, then per month at the stated volume (a month is 30.4 days), using the table above and no caching unless stated.

Example	Tokens in / out	Volume	Haiku 5.5	Sonnet 5.5	Opus 5.5
Chat assistant	1,500 / 400	100 messages a day	$0.00035 → $1.06/mo	$0.007 → $21.28/mo	$0.014 → $42.56/mo
Document summariser	10,000 / 500	100 documents a day	$0.00125 → $3.80/mo	$0.025 → $76/mo	$0.05 → $152/mo
Classifier or router	500 / 20	1,000 items a day	$0.00006 → $1.82/mo	$0.0012 → $36.48/mo	$0.0024 → $72.96/mo

Chat assistant. 1,500 input tokens covers short instructions plus a few turns of conversation; 400 output is a paragraph or two. On Sonnet 5.5 that is $0.007 a message. If half of the input is a fixed instruction block served from cache, the message drops to $0.0056 and the month to about $16.95.

Document summariser. Input dominates: 10,000 tokens of document against 500 of summary. The document is different every time, so caching does not help, but the Batch API does: run it overnight and the Sonnet 5.5 bill halves to $38 a month.

Classifier. Twenty output tokens is a label and a confidence. Haiku 5.5 is the natural choice at $1.82 a month for 30,400 items; Sonnet 5.5 costs twenty times more for a task it is rarely needed for.

Put your own numbers into the AI app cost calculator to see the API line next to hosting, database and email.

Pay-per-token or a $20 plan?

Claude Pro costs $20 a month ($17 on annual billing) and includes the apps, file handling, web search and usage limits. The API has no limits but no apps either. For the same usage, which is cheaper?

Thirty messages a day of 3,000 input and 600 output tokens (the input includes the conversation so far, which is why it is larger than a question):

Model	Per message	Per month	Break-even vs Pro
Claude Haiku 5.5	$0.0006	$0.55	about 1,100 messages a day
Claude Sonnet 5.5	$0.012	$10.94	about 55 messages a day
Claude Opus 5.5	$0.024	$21.89	about 27 messages a day
Claude Fable 5.1	$0.06	$54.72	about 11 messages a day

So a Sonnet-class model through the API is cheaper than Pro until you send about 55 long messages a day, and Opus 5.5 is already more expensive than Pro at 30. The subscription vs API calculator takes your own message count and length.

What moves the bill

Context length. Every message re-sends the conversation. A 40-message session sends the first message 40 times. Trim history, summarise old turns, or start new sessions.
Model choice. Haiku 5.5 to Sonnet 5.5 is 20×; Sonnet 5.5 to Opus 5.5 is 2×; Opus 5.5 to Fable 5.1 is 2.5×. Route the easy requests to the cheap model and keep the expensive one for the hard ones.
Caching and batching. A cached instruction block costs 5% of its normal price on Sonnet 5.5; the Batch API halves everything that can wait. Together they often cut a bill by more than half without touching quality.

What the token price leaves out

The model is one line of the bill. A small Claude-based tool also needs somewhere to run (a $5 server is enough for most of ours, see the ledger), possibly a database and an email service, and your time. The AI app cost calculator adds those lines; if you would rather someone else built and ran it, the build page shows three example tools with their full monthly running cost.

Questions

Is the Claude API expensive?

For light and medium use, no. A 3,000-token question with a 600-token answer costs about 1.2 cents on Claude Sonnet 5.5 and 0.06 cents on Claude Haiku 5.5. The bill becomes noticeable with long conversations, large documents in every request, or thousands of requests a day, and that is where caching, a cheaper model or the Batch API make the difference.

How much does one million tokens cost on Claude?

At list prices checked on 28 October 2026: Claude Haiku 5.5 $0.10 in and $0.50 out, Claude Sonnet 5.5 $2 in and $10 out, Claude Opus 5.5 $4 in and $20 out, Claude Fable 5.1 $10 in and $50 out. One million tokens is roughly 750,000 English words.

Is the Claude API cheaper than Claude Pro?

Up to a point. Thirty messages a day of 3,000 input and 600 output tokens cost about $11 a month on Sonnet 5.5 against $20 for Pro; the API stays cheaper until about 55 such messages a day. On Opus 5.5 the break-even is about 27 messages a day and on Fable 5.1 about 11. The plan also includes the apps and features the API does not.

Sources

Prices in this guide were checked on the date shown above and change often. Check each provider's pricing page before you commit.