Claude API cost calculator

Anthropic list prices as of October 30, 2026

Describe one typical request and how many you send, and the calculator works out the monthly cost on every current Claude model, with prompt caching and without it. It uses Anthropic's API list prices as of October 30, 2026, and everything runs in your browser.

Requests per month Calls to the Messages API in a month.

Cached prompt tokens per request The part that repeats from call to call: system prompt, tool definitions, documents and earlier turns.

New input tokens per request The part that changes each call and isn't read from the cache.

Output tokens per request What the model writes back, thinking included.

Cache hit rate, % Share of requests that read the cached prompt; the rest write it.

Cache duration A five-minute write costs 1.25 times the input price, a one-hour write twice it.

Batch API Half price on every token for asynchronous jobs.

US-only inference 1.1 times every token price, with inference_geo: "us".

Monthly cost by model

Model	Without caching	With caching	Caching saves
Claude Fable 5.1	$0.00	$0.00	$0.00
Claude Opus 5.5	$0.00	$0.00	$0.00
Claude Sonnet 5.5	$0.00	$0.00	$0.00
Claude Haiku 5.5	$0.00	$0.00	$0.00

Claude Haiku 5.5 uses its prices for prompts up to 100,000 tokens.

Prices used

These are Anthropic's prices in US dollars per million tokens as of October 30, 2026, from its pricing page. The cache columns are what it costs to write the prompt to the cache for five minutes or an hour, and to read it back.

Model	Input	Output	5-minute write	1-hour write	Cache read
Claude Fable 5.1	$10	$50	$12.50	$20	$0.25
Claude Opus 5.5	$4	$20	$5	$8	$0.20
Claude Sonnet 5.5	$2	$10	$2.50	$4	$0.10
Claude Haiku 5.5, prompts up to 100,000 tokens	$0.10	$0.50	$0.125	$0.20	$0.01
Claude Haiku 5.5, prompts over 100,000 tokens	$0.50	$2.50	$0.625	$1	$0.05

How is the cost worked out?

Each request is priced in three parts and multiplied by the number of requests a month. Without caching, the cached and new input tokens are both billed at the input price. With caching, the share of requests that miss the cache write the cached tokens at the write price, and the share that hit read them at the cache-read price.

Cached prompt: hits times the cache-read price, plus misses times the five-minute or one-hour write price.
New input: every request's new tokens at the input price.
Output: every request's output tokens at the output price.

The Batch API halves every price and US-only inference multiplies every price by 1.1, and both stack with the caching prices, as Anthropic's pricing page describes. Tool definitions, images and thinking count as the input or output tokens they become; server-side tools such as web search are billed apart and aren't included.

When does prompt caching pay off?

Prompt caching pays for itself after one cache read with the five-minute cache and after two with the one-hour cache, on every current Claude model. A five-minute write costs a quarter more than plain input, and the first read saves at least nine tenths of the input price; a one-hour write costs twice the input price, so it takes a second read to come out ahead. A cache that keeps missing costs more than no cache at all, which the calculator shows when the hit rate is low.

Choose the one-hour cache when requests sharing a prompt arrive more than five minutes apart, such as a session someone comes back to. Anthropic's prompt caching guide covers where to place cache breakpoints and the minimum length a cached prompt must have.

What does caching cost in Claude Code?

Writing the prompt cache was the largest part of Claude Code's cost in AutoRouter's October 2026 measurements: 63% of the list-price bill on Claude Opus 5.5, 66% on Claude Sonnet 5.5 and 51% on Claude Haiku 5.5, across 45 runs of five everyday tasks. Each fresh session writes Claude Code's system prompt, its tool definitions and the files it reads into the cache, so short sessions pay that setup every time.

The Claude Code cost calculator estimates a month of those tasks on each model, and what common Claude Code tasks cost, measured on each model has the full study. AutoRouter, the Mac app for Codex and Claude Code, sends each task to the model that does it best, so quick work doesn't run on the most expensive model.

Questions

Does the Batch API discount stack with prompt caching?

Yes. Anthropic's pricing page says the caching multipliers stack with other pricing modifiers, including the Batch API discount and data residency, so a batched cache read costs half the usual cache-read price.

Why are cache reads cheaper on Claude Opus 5.5 and Sonnet 5.5?

As of October 30, 2026, a cache hit costs 5% of the input price on Claude Opus 5.5 and Claude Sonnet 5.5, 2.5% on Claude Fable 5.1 and 10% on other models such as Claude Haiku 5.5. The write prices follow the usual 1.25 and 2 times multipliers on every model.

Does Claude Haiku 5.5 cost more for long prompts?

Yes. A Claude Haiku 5.5 request whose prompt is over 100,000 tokens pays five times the shorter-prompt prices, and the prompt length counts cache reads and cache writes as well as new input. The calculator switches to those prices when the cached and new input tokens together pass 100,000.

Is this what Claude Code costs on a Claude plan?

No. The calculator uses API list prices, which apply to an API key. On a Claude Pro or Max plan, Claude Code uses the plan's limits instead of money; the measured cost of everyday Claude Code tasks is on the Claude Code cost calculator.

Sources

Pricing, Claude Platform docs, Anthropic, read October 30, 2026
Prompt caching, Claude Platform docs, Anthropic
AutoRouter research: measured Claude Code costs, CC BY 4.0