How much does Claude OCR cost?

Anthropic does not publish a per-page OCR rate, only per-million-token prices, so you have to convert. On a scanned US Letter page at 150 DPI, Claude Opus 4.8 works out to about $32 per 1,000 pages, Claude Sonnet 5 about $19, and Claude Haiku 4.5 about $5.31. Those figures include both the image tokens going in and roughly 750 transcription tokens coming out, which is the part most cost estimates leave off.

How much does Claude cost per page for OCR?

About $0.0053 per page on Claude Haiku 4.5, $0.019 on Claude Sonnet 5, and $0.032 on Claude Opus 4.8, for a scanned letter-size page transcribed to text. Multiply by 1,000 for the per-1,000-pages figure. The number moves with page density, because a page with more text produces more output tokens, and output is where most of the money goes.

How many tokens does a Claude PDF page use?

Two separate amounts, and they stack. Anthropic states that extracted text runs 1,500 to 3,000 tokens per page depending on density, and that each page is additionally converted into an image and billed at image rates. A scanned letter page at 150 DPI is 2,714 visual tokens on the high-resolution models and about 1,560 on the standard-resolution ones. A born-digital PDF pays both layers.

How does Claude calculate image tokens?

Claude views an image as 28 by 28 pixel patches, so an image costs the ceiling of width divided by 28, times the ceiling of height divided by 28, in visual tokens. A 1,000 by 1,000 pixel image is 36 by 36, which is 1,296 tokens. The older formula of width times height divided by 750, still repeated in many guides, is retired and gives the wrong answer.

Why is Claude Haiku so much cheaper for OCR?

Partly the token price, and partly because Haiku 4.5 sits on the standard resolution tier, capped at a 1,568 pixel long edge and 1,568 visual tokens. A letter-size page that costs 2,714 visual tokens on Opus 4.8 is downscaled to roughly 1,560 on Haiku, a 42 percent cut. You pay less because Claude is looking at a smaller version of the page, which matters most on small type and dense tables.

Is Claude cheaper than AWS Textract for OCR?

No. AWS Textract, Azure AI Document Intelligence and Google Document AI all charge $1.50 per 1,000 pages for plain text OCR. The cheapest Claude path, Haiku 4.5, is about $5.31 per 1,000 pages, roughly 3.5 times more, and Opus 4.8 at about $32 is roughly 21 times more. Claude is not competing on OCR price. It is competing on reading documents that classical OCR mangles.

How many pages can Claude read in one PDF?

Up to 600 pages per request on the models with a 1 million token context window, and 100 pages when the context window is under 1 million, which is where Claude Haiku 4.5 sits. The whole request also has to stay inside 32 MB. The 100-page figure repeated in most write-ups is the smaller-context number, not a universal Claude limit.

Do output tokens really matter for OCR pricing?

They dominate it. Transcribing a page produces roughly 750 output tokens, billed at five times the input rate on every Claude model. On Opus 4.8 that is $18.75 of a $32.32 bill per 1,000 pages, about 58 percent. On Haiku 4.5 it is $3.75 of $5.31, about 71 percent. Any estimate built on the input price alone understates the real cost by roughly two and a half to three and a half times.

Which Claude model is best for document extraction?

Sonnet 5 is the usual answer: it is on the high-resolution tier, so it sees the full page, at about $19 per 1,000 pages, or roughly $13 while the introductory rate runs to August 31, 2026. Haiku 4.5 is the budget option at about $5.31 but downscales dense pages. Opus 4.8 is worth its roughly $32 only on documents where the reasoning genuinely changes the answer, such as messy handwriting or complex nested tables.

Does Claude charge extra for PDFs?

There is no separate PDF fee. Standard token pricing applies, but a PDF page is billed on two meters at once: the text Claude extracts from it, and the image Claude renders it as. Anthropic's own Bedrock documentation quantifies the gap, at roughly 1,000 tokens for a 3-page PDF in text-only mode against roughly 7,000 tokens for the same 3 pages with full visual understanding, a factor of seven.

// Verified rate reference, July 22, 2026

Claude OCR Pricing: Cost Per Page and Per 1,000 Pages

Anthropic publishes token prices, not page prices, so nobody quotes what Claude OCR actually costs. Converted, a scanned letter page runs $5.31 per 1,000 pages on Haiku 4.5, $19.39 on Sonnet 5 and $32.32 on Opus 4.8. Here is the full arithmetic, including the output tokens most estimates quietly drop.

Every model in dollars per 1,000 pages
The 28-pixel patch formula, verified
Why output tokens are most of the bill
The resolution-tier trap

See every rate Extract a document free

Upload a document, no signup

PDF, JPG, PNG, BMP, HEIC, TIFF

Upload a document to extract

Extract documents →

A token price says nothing about whether the fields come out right on your documents. Drop one in and check the output before you commit.

SOC 2 Type II

256-bit encryption

US data handling

Fields, not just text

$5.31: Haiku 4.5, per 1,000 pages
$32.32: Opus 4.8, per 1,000 pages
58%: of the Opus bill is output tokens
2,714: visual tokens in a letter page

// The short answer

What Claude OCR costs

Anthropic sells tokens, not pages, so there is no official Claude OCR rate to look up. Converting a scanned US Letter page at 150 DPI, Claude Haiku 4.5 costs about $5.31 per 1,000 pages, Claude Sonnet 4.6 about $15.93, Claude Sonnet 5 about $19.39 (roughly $12.93 while the introductory rate runs to August 31, 2026), Claude Opus 4.8 and 4.7 about $32.32, and Claude Fable 5 about $64.64. Those totals include the image tokens going in and roughly 750 transcription tokens coming out, which is where most of the money actually goes. A born-digital PDF costs more again, because Anthropic bills the extracted text layer of 1,500 to 3,000 tokens per page on top of the page image. Even the cheapest Claude path is about 3.5 times plain cloud OCR at $1.50 per 1,000 pages, and Opus 4.8 is about 21 times. You are not buying cheaper OCR, you are buying a model that reasons about the page.

// Every model, converted to dollars

Claude OCR pricing per 1,000 pages

One scanned US Letter page at 150 DPI, which is 1,275 by 1,650 pixels, transcribed to text at roughly 750 output tokens. Those two assumptions are stated so you can redo the math for your own documents.

Model	Tier	Visual tokens / page	Input / 1,000 pages	Output / 1,000 pages	Total / 1,000 pages
Claude Fable 5	High-res	2,714	$27.14	$37.50	$64.64
Claude Opus 4.8	High-res	2,714	$13.57	$18.75	$32.32
Claude Opus 4.7	High-res	2,714	$13.57	$18.75	$32.32
Claude Sonnet 5 (list)	High-res	2,714	$8.14	$11.25	$19.39
Claude Sonnet 5 (intro, to Aug 31 2026)	High-res	2,714	$5.43	$7.50	$12.93
Claude Sonnet 4.6	Standard	~1,560	$4.68	$11.25	$15.93
Claude Haiku 4.5	Standard	~1,560	$1.56	$3.75	$5.31

Derived on July 22, 2026 from Anthropic's published per-million-token prices and the visual-token rules in its vision documentation. A born-digital PDF adds an extracted-text layer of 1,500 to 3,000 tokens per page on top of these figures. Re-verify before quoting.

// The published rates

Claude API pricing per million tokens

These are Anthropic's own numbers, unconverted. Everything above is built from them.

Model	Input / 1M tokens	Output / 1M tokens	Context	Max PDF pages / request
Claude Fable 5	$10.00	$50.00	1M	600
Claude Opus 4.8	$5.00	$25.00	1M	600
Claude Opus 4.7	$5.00	$25.00	1M	600
Claude Sonnet 5	$3.00 ($2.00 intro)	$15.00 ($10.00 intro)	1M	600
Claude Sonnet 4.6	$3.00	$15.00	1M	600
Claude Haiku 4.5	$1.00	$5.00	200K	100

Output is billed at five times input on every current Claude model. That single ratio is why transcription, which is output-heavy, costs more than the input price suggests.

// Read this before you budget

Four things that change the real cost

The token prices are simple. The rules that turn them into a page price are where the published estimates go wrong.

Output tokens are most of the bill

Transcribing a page produces text, and output costs five times input on every Claude model. At roughly 750 output tokens per page, that is $18.75 of a $32.32 total on Opus 4.8, about 58 percent, and $3.75 of $5.31 on Haiku 4.5, about 71 percent. Estimates built on the input price alone land two and a half to three and a half times low. If you only need a handful of fields rather than the whole page, ask for just those fields, because shorter output is the single biggest lever you control.

The resolution-tier trap

Claude has two image tiers. Fable 5, Opus 4.8, Opus 4.7 and Sonnet 5 allow a 2,576 pixel long edge and 4,784 visual tokens. Every other model, including Haiku 4.5 and Sonnet 4.6, caps at 1,568 pixels and 1,568 visual tokens. A letter page that costs 2,714 visual tokens on Opus is downscaled to about 1,560 on Haiku, a 42 percent cut. The cheap models are partly cheap because Claude is looking at a smaller page, and that is exactly where small type and dense tables start to fail.

A PDF page is billed twice

Anthropic converts every PDF page into an image and also extracts its text, then bills both. The documentation puts the text layer at 1,500 to 3,000 tokens per page depending on density, with image token costs applied on top. Anthropic's own Bedrock figures show the size of the gap: roughly 1,000 tokens for a 3-page PDF in text-extraction-only mode against roughly 7,000 tokens for the same 3 pages with full visual understanding. That is seven times, on the same file.

The page limit is a context limit

Most write-ups say Claude reads 100 pages per request. That is the number for models whose context window is under 1 million tokens, which today means Haiku 4.5. On the 1 million token models the limit is 600 pages, the highest single-request page ceiling of any document API we track, ahead of Azure Document Intelligence at 2,000 pages but on a much larger token budget. The 32 MB request cap usually bites first, which is why large files should go through the Files API and be referenced by id.

// Do the math yourself

How Claude counts image tokens

Claude reads an image as a grid of 28 by 28 pixel patches. Each patch is one visual token.

visual tokens = ceil(width / 28) x ceil(height / 28)

A scanned US Letter page at 150 DPI is 1,275 by 1,650 pixels. That is 46 patches across and 59 down, so 2,714 visual tokens, which fits inside the high-resolution budget of 4,784 without any downscaling. On a standard-tier model the same page breaks both the 1,568 pixel long edge and the 1,568 token budget, so it is scaled down to roughly 1,560 tokens before Claude ever sees it.

This formula replaced the older width times height divided by 750 rule, which is still repeated across the web and gives noticeably wrong answers. We checked the current formula against three values Anthropic publishes itself: a 200 by 200 image comes out at 8 times 8, which is 64 tokens; a 1,000 by 1,000 image at 36 times 36, which is 1,296; and a 1,092 by 1,092 image at 39 times 39, which is 1,521. All three match the published table exactly.

42%

fewer visual tokens on the cheap tier

2,714 tokens on Opus 4.8 and Sonnet 5, about 1,560 on Haiku 4.5 and Sonnet 4.6, for the identical page. Lower cost, smaller picture. Raising your scan DPI does not help on a standard-tier model, because it just gets scaled back down.

// In context

Claude versus the dedicated OCR APIs

Claude is a reasoning model that happens to read pages, not an OCR meter, and the gap shows. For the full cross-vendor table see OCR pricing per 1,000 pages, or the sibling reference on Gemini OCR pricing.

Service	Per 1,000 pages	What you get	Note
Claude Haiku 4.5	$5.31	Reasoning over the page	Downscales dense pages
Claude Sonnet 5	$19.39	Reasoning, full resolution	$12.93 on the intro rate
Claude Opus 4.8	$32.32	Strongest reasoning	The premium tier
AWS Textract (Detect Document Text)	$1.50	Raw text and boxes	Drops to $0.60 above 1M pages
Azure AI Document Intelligence (Read)	$1.50	Raw text and boxes	Commitment tiers to $0.45
Google Document AI (Enterprise OCR)	$1.50	Raw text and boxes	Drops to $0.60 above 5M pages
Mistral OCR 4	$4.00	Markdown, structured	Batch mode halves it
LlamaParse Agentic	$12.50	Markdown, tables, reasoning	Cheapest true agentic parse

Where Claude earns the premium: handwriting, unusual layouts, documents where the right answer depends on reading the page rather than matching a template. Where it does not: clean, high-volume, repetitive forms, which is what the $1.50 services were built for. Most teams end up routing rather than picking one.

Frequently asked questions

How much does Claude OCR cost?: Anthropic does not publish a per-page OCR rate, only per-million-token prices, so you have to convert. On a scanned US Letter page at 150 DPI, Claude Opus 4.8 works out to about $32 per 1,000 pages, Claude Sonnet 5 about $19, and Claude Haiku 4.5 about $5.31. Those figures include both the image tokens going in and roughly 750 transcription tokens coming out, which is the part most cost estimates leave off.
How much does Claude cost per page for OCR?: About $0.0053 per page on Claude Haiku 4.5, $0.019 on Claude Sonnet 5, and $0.032 on Claude Opus 4.8, for a scanned letter-size page transcribed to text. Multiply by 1,000 for the per-1,000-pages figure. The number moves with page density, because a page with more text produces more output tokens, and output is where most of the money goes.
How many tokens does a Claude PDF page use?: Two separate amounts, and they stack. Anthropic states that extracted text runs 1,500 to 3,000 tokens per page depending on density, and that each page is additionally converted into an image and billed at image rates. A scanned letter page at 150 DPI is 2,714 visual tokens on the high-resolution models and about 1,560 on the standard-resolution ones. A born-digital PDF pays both layers.
How does Claude calculate image tokens?: Claude views an image as 28 by 28 pixel patches, so an image costs the ceiling of width divided by 28, times the ceiling of height divided by 28, in visual tokens. A 1,000 by 1,000 pixel image is 36 by 36, which is 1,296 tokens. The older formula of width times height divided by 750, still repeated in many guides, is retired and gives the wrong answer.
Why is Claude Haiku so much cheaper for OCR?: Partly the token price, and partly because Haiku 4.5 sits on the standard resolution tier, capped at a 1,568 pixel long edge and 1,568 visual tokens. A letter-size page that costs 2,714 visual tokens on Opus 4.8 is downscaled to roughly 1,560 on Haiku, a 42 percent cut. You pay less because Claude is looking at a smaller version of the page, which matters most on small type and dense tables.
Is Claude cheaper than AWS Textract for OCR?: No. AWS Textract, Azure AI Document Intelligence and Google Document AI all charge $1.50 per 1,000 pages for plain text OCR. The cheapest Claude path, Haiku 4.5, is about $5.31 per 1,000 pages, roughly 3.5 times more, and Opus 4.8 at about $32 is roughly 21 times more. Claude is not competing on OCR price. It is competing on reading documents that classical OCR mangles.
How many pages can Claude read in one PDF?: Up to 600 pages per request on the models with a 1 million token context window, and 100 pages when the context window is under 1 million, which is where Claude Haiku 4.5 sits. The whole request also has to stay inside 32 MB. The 100-page figure repeated in most write-ups is the smaller-context number, not a universal Claude limit.
Do output tokens really matter for OCR pricing?: They dominate it. Transcribing a page produces roughly 750 output tokens, billed at five times the input rate on every Claude model. On Opus 4.8 that is $18.75 of a $32.32 bill per 1,000 pages, about 58 percent. On Haiku 4.5 it is $3.75 of $5.31, about 71 percent. Any estimate built on the input price alone understates the real cost by roughly two and a half to three and a half times.
Which Claude model is best for document extraction?: Sonnet 5 is the usual answer: it is on the high-resolution tier, so it sees the full page, at about $19 per 1,000 pages, or roughly $13 while the introductory rate runs to August 31, 2026. Haiku 4.5 is the budget option at about $5.31 but downscales dense pages. Opus 4.8 is worth its roughly $32 only on documents where the reasoning genuinely changes the answer, such as messy handwriting or complex nested tables.
Does Claude charge extra for PDFs?: There is no separate PDF fee. Standard token pricing applies, but a PDF page is billed on two meters at once: the text Claude extracts from it, and the image Claude renders it as. Anthropic's own Bedrock documentation quantifies the gap, at roughly 1,000 tokens for a 3-page PDF in text-only mode against roughly 7,000 tokens for the same 3 pages with full visual understanding, a factor of seven.

Test the output on your own documents

A token price tells you nothing about accuracy. Upload a document to DocuOCR, see the extracted fields, and compare before you build a pipeline around any model.

Extract a document free Compare every OCR rate

Related pricing references

Claude OCR

How Claude reads documents, and where it fails

Gemini OCR pricing

Token rates converted to dollars per 1,000 pages

LLM OCR

When a language model beats classical OCR

Mistral OCR pricing

$4 per 1,000 pages, and half that in batch

OCR pricing per 1,000 pages

Every vendor, one rate table

Agentic document extraction pricing

LlamaParse, Reducto, Landing AI, Extend

From the same family of tools