Google Document AI Invoice Parser Pricing (2026)

Last updated July 2026.

Google Document AI charges $0.10 for every 10 pages in a document for the prebuilt Invoice Parser. The catch is that the block is the billing unit, so a one-page invoice is billed exactly the same as a ten-page invoice: $0.10. Because most invoices are one or two pages, the real cost of parsing 1,000 typical invoices is around $100, not the $10 per 1,000 pages the block rate seems to imply. That is ten times what AWS Textract Analyze Expense charges for the same job, and ten times Azure's prebuilt invoice model.

This is the single most misunderstood rate in document AI pricing, and it is misunderstood because Google's own page states it accurately while almost every third-party guide converts it wrongly. The rate is real, the arithmetic is not hard, and the difference between getting it right and getting it wrong is a factor of ten on your invoice processing bill.

How much does Google Document AI Invoice Parser cost?

Straight from Google's pricing page, verified July 2026:

Google words it as "$0.10 for every 10 pages in a document". Read that carefully. It is not $0.01 per page. It is $0.10 per block of up to ten pages, and a partial block bills as a full one.

The same block pricing applies to the other prebuilt processors in that family, including the Expense Parser and the Utility Parser.

Why a one-page invoice costs $0.10

Here is where the money goes. An invoice is a document. A one-page invoice is a one-page document, which falls in the 1-to-10-page band, which costs $0.10.

So:

Every row costs the same $100, because you sent 1,000 documents and each one consumed one block. The effective per-page rate swings by a factor of ten purely on how long your invoices happen to be. And since the overwhelming majority of business invoices are one or two pages, most teams are living in the top row and quoting themselves the bottom one.

Can you avoid the block charge by merging invoices?

No, and this is the trap inside the trap. The obvious dodge is to staple ten single-page invoices into one ten-page PDF, pay a single $0.10 block, and save 90%. It does not work.

The Invoice Parser extracts one invoice's fields from a document. Give it a ten-page PDF containing ten different invoices and it does not return ten records. It returns one confused record, with the vendor from one page, the total from another, and line items blended from all of them. You have not saved $0.90; you have destroyed your data.

The block discount is real, but it only rewards documents that are genuinely long. A ten-page invoice from a construction supplier with pages of line items gets parsed for $0.10, and that is a genuine bargain. A one-page invoice from your landlord costs the same $0.10 and there is nothing you can do about it. Budget for the invoices you actually receive, not for the ones the pricing model would prefer you to receive.

How does that compare to Textract and Azure?

This is where the block unit really bites, because the other two hyperscalers bill invoices per page with no block rounding.

For the single-page invoices that make up most of real accounts payable work, Google's prebuilt Invoice Parser is the most expensive option of the four, by a factor of ten against the two obvious rivals. That is not a knock on its accuracy, which is good. It is a statement about the billing unit, and the billing unit is the thing that decides what you pay.

If invoices are the bulk of your volume, the practical move is either to use a different vendor's prebuilt model, or to use Google's own Custom Extractor instead, which bills at a flat $30 per 1,000 pages with no block rounding. On 1,000 single-page invoices that is $30 against the Invoice Parser's $100. You give up the prebuilt schema and train your own, which is real work, but the arithmetic is hard to argue with at volume.

What is Google Document AI pricing per page?

Since people search for this directly, here is the rest of the rate card, verified from Google's pricing page in July 2026. Note that only the prebuilt parsers use block billing. Everything else is a straightforward per-1,000-pages rate.

The hosting fee that accrues while you sleep

One more line that catches people, and it has nothing to do with pages. If you deploy a custom processor version, Google charges $0.05 per hour per deployed version, whether or not you send it a single document. Left running for a year that is $438 per version, and it is easy to leave two or three versions deployed after testing.

That fee is invisible in any per-page comparison, and it is the reason a low-volume custom pipeline on Google can cost more than a high-volume one on a service that does not charge for idle capacity. Undeploy the versions you are not using.

On the credit side, Google does something none of its rivals do: failed requests are not billed. A 4xx or 5xx response costs you nothing. AWS and Azure both bill some failure modes.

What should you actually do about invoices?

Work out the average page count of the invoices you really receive, then price the job three ways: Google's Invoice Parser at $0.10 a block, Google's Custom Extractor at $30 per 1,000 pages, and a rival prebuilt model at $10 per 1,000 pages. For most accounts payable teams, whose invoices are one page and arrive by the thousand, the block unit makes Google's prebuilt parser the wrong default, and almost nobody notices because the headline number looks small.

And if the actual goal is to get invoice data out of PDFs and into something a person can reconcile, rather than to run an extraction API at all, you can pull the invoice data straight into a spreadsheet without writing any code against a cloud processor. Plenty of finance teams reach for an API when a converter would have finished the job by lunchtime.

Whichever route you pick, price the unit, not the headline. Per page, per document and per 10-page block are three different things, and the vendors are not lining them up for you.

Related reading: Google Document AI pricing in full, OCR pricing per 1,000 pages across every vendor normalized onto one unit, per page vs per document billing, and the invoice OCR money page if you want the fields extracted and reviewed rather than just parsed.