Guides / claude haiku 4.5 pricing

Claude Haiku 4.5 pricing at a million calls a month

Claude Haiku 4.5 pricing is $1.00 per million input tokens and $5.00 per million output tokens, with cached reads at $0.10 per million. It runs at call volumes where a small per-call regression becomes a large monthly number. Culpa, a local-first LLM cost, margin, and forecast ledger, meters every call so a high-volume tier can't drift unnoticed.

Why this happens

Haiku sits at the high-frequency end of the roster, handling classification, routing and the cheap first pass in a two-model chain. That volume changes which mistakes matter. A hundred extra prompt tokens is irrelevant on a frontier model at low volume and a four-figure monthly line here. The cached rate is a tenfold discount, which is the largest proportional saving available anywhere in this price book.

What this usually looks like

  • Haiku carries most of your call count and a surprising share of your spend.
  • Cached-token share is near zero, which at this volume is the most expensive miss you can make.
  • Nobody reviews the Haiku prompt because the per-call cost rounds to nothing.
  • A two-model chain sends more to Haiku than the expensive model it was meant to protect.

Free, no card, no account

Run the free Cost Leak Scan

It shows your most expensive conversation before you install anything.

Run a free scan

Mistakes that cost the most

MistakeWhy it hurtsDo instead
Leaving the cached rate unused at high call volume.The cached rate is a tenth of input, so at millions of calls the untaken discount is the largest line you can recover.Move the stable block to the front of the prompt and check the cached share weekly.
Rebuilding the cache more often than you read it.The five-minute cache write bills at $1.25 per million, above the $1.00 input rate.Track writes against reads. A five-minute cache clears break-even on the first read, so only a cache rebuilt without ever being read loses money.
Skipping prompt review because the rate is low.At a million calls, every thousand prompt tokens is $1,000 a month.Price the prompt explicitly: calls times tokens times $1.00 per million. Then decide if it earns its length.

Run this check tonight

  1. Multiply your monthly Haiku call count by prompt tokens by $1.00 per million. That's your fixed prompt cost.
  2. Divide cached input tokens by total input tokens. At this volume, a low hit rate is your biggest single loss.
  3. Count cache writes against reads. More writes than reads means the cache is costing you.
  4. Check what share of Haiku output is boilerplate you could have templated outside the model.
  5. Confirm Haiku is handling mechanical work rather than work it fails at and escalates.

One million calls, cached and uncached

Illustrative example

A classification step running 1 million calls a month with a 4,000-token system prompt and 80 tokens of output. Rates from the price book: $1.00 input, $5.00 output, $0.10 cached read, $1.25 five-minute cache write.

Uncached input: 1,000,000 x 4,000 = 4,000M tokens x $1.00 per million = $4,000
Output: 1,000,000 x 80 = 80M tokens x $5.00 per million = $400
Uncached month total = $4,400
Cached input at an 80% hit rate: (3,200M x $0.10) + (800M x $1.00) = $320 + $800 = $1,120
Cached month total = $1,520, a saving of $2,880 from prompt ordering alone

$2,880 a month sits in the difference between a working cache and a broken one, on a model everyone describes as cheap. The output side is under a tenth of the bill.

Every number, with its confidence and source

FigureWhat it meansConfidenceSource
$1.00 / $5.00 per millionClaude Haiku 4.5 input and output ratescalculatedCulpa price book row for anthropic/claude-haiku-4-5, $0.001 and $0.005 per 1k, effective 2026-07-02, verified against Anthropic pricing.
10xcost of uncached input against the cached rate on Haiku 4.5calculated$1.00 input divided by the $0.10 cached rate, from the same price-book row.
$1,520 to $4,400modelled monthly cost of a one-million-call classification step, cached versus uncachedestimatedBoth endpoints from the teardown arithmetic at real Haiku 4.5 rates. A range because call volume and prompt size are modelled.

What a generic answer can’t know

A rate card can't tell you that 91% of your Haiku input is a system prompt, or that your cached share collapsed after a deploy edited it. That needs prompt structure compared across calls. Culpa reads it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.

Questions founders ask next

How much does Claude Haiku 4.5 cost?

$1.00 per million input tokens and $5.00 per million output, verified 2026-07-02 against Anthropic's pricing page. Cached reads are $0.10 per million, a tenfold discount that matters more here than on any other tier because of the call volume.

Is the dated Haiku model a different price?

No. The dated alias in this price book bills at exactly the same rates as the base model, so pinning the dated version costs nothing and changes nothing about your bill.

Why does caching matter so much on Haiku?

Because the discount is proportionally the same as elsewhere but the call volume is far higher. A tenfold saving on four thousand repeated tokens across a million calls is thousands of dollars a month, from one change to prompt ordering.

On your infrastructure

Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.

Free, no card, no account

Run the free Cost Leak Scan

It shows your most expensive conversation before you install anything.

Run a free scan

Keep reading


Sources: Anthropic pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.