Google Document AI costs $1.50 per 1,000 pages for Enterprise Document OCR, with the first 1,000 pages of each billing cycle free. Layout Parser is $10.00 per 1,000 pages, and Form Parser or a custom extractor is $30.00. The pretrained parsers are the trap: they bill per count, and Google defines one count as up to ten pages, so a one-page invoice costs the same as a ten-page one.
Written for US teams costing out a document API. Every rate here was read off Google Cloud's published rate card on 21 August 2026, not retyped from memory. Last updated August 2026.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Free plan extracts the first 5 files, the rest unlock when you upgrade
Uploading...
Before you pick a Document AI processor, drop in the document you meant to send it and see what finished, validated fields look like.
Reading a page costs $1.50 per 1,000 with Enterprise Document OCR, and that is the number everyone quotes. Understanding the page is where the money goes: Layout Parser is $10.00 per 1,000 pages and Form Parser is $30.00, a twenty times jump off the headline. But the thing that actually surprises finance teams is not a rate at all, it is a unit. Google bills four different ways inside one product, and the pretrained parsers round up to ten-page blocks, so the same $0.10 buys you one page or ten. Work out which unit your documents fall into before you compare Google's rate card with anyone else's, because a per-page number for Document AI is only true for half the processors on it.
These are the processors that bill per page, so a per-1,000-pages figure is meaningful for all of them. Google confirms the unit in its own worked example on the same page: 100 pages through Form Parser is calculated as ($30 / 1,000 pages) times 100 pages, which comes to $3.
| What you are doing | Processor | Per 1,000 pages | Above the break | Break at | Note |
|---|---|---|---|---|---|
| Plain text and layout OCR | Enterprise Document OCR |
$1.50 | $0.60 | 5,000,000 pages | First 1,000 pages $0.00 |
| Optional OCR extras (v2 only) | OCR add-ons |
$6.00 | no break published | n/a | On top of the base rate |
| Structure, chunks and headings | Layout Parser (includes initial chunking) |
$10.00 | no break published | n/a | Chunking included |
| Re-chunking an already parsed document | Re-chunking |
$0.02 | no break published | n/a | Priced per 1,000 count |
| Split a batch into documents | Custom splitter |
$5.00 | $3.00 | 1,000,000 pages | You train it |
| Sort documents into classes | Custom classifier |
$5.00 | $3.00 | 1,000,000 pages | You train it |
| Key-value pairs and tables, untrained | Form Parser |
$30.00 | $20.00 | 1,000,000 pages | No training needed |
| Your own fields, trained | Custom extractor |
$30.00 | $20.00 | 1,000,000 pages | Plus $0.05 per hour hosting |
Read off Google Cloud's published Document AI rate card on 21 August 2026. Google's Cloud Billing Catalog API needs an API key and the old public price list JSON now returns 404, so the pricing page is the primary source here. One row is deliberately missing: Google lists a Summarizer at $0.025 with no unit attached, and since we cannot establish whether that is per count or per 1,000, we have left it off rather than guess.
This is easy to miss and expensive to miss. Enterprise Document OCR holds $1.50 per 1,000 pages all the way to five million pages a month before dropping to $0.60. Form Parser, custom extractors, the custom splitter and the custom classifier all break at one million. AWS Textract and Azure Document Intelligence both break their text meters at one million too, so a team running between one and five million pages a month is paying Google's full rate while the same volume elsewhere has already been discounted. Google also bills these tiers as a graduated scale and shows the arithmetic: six million pages through Form Parser is $30,000 for the first million plus $100,000 for the next five, not $120,000 flat.
A custom extractor has two lines on your bill, and only one of them is prediction. Hosting is $0.05 per hour per deployed processor version, which Google's own example works out to $438 for a year. Nothing about that charge depends on traffic, so a version you deployed to test in March and never undeployed has quietly cost you the same as one running in production. Teams that version aggressively end up with several deployed at once. Three idle versions is $1,314 a year for output nobody reads. The provisioned service tier follows the same logic at $0.410958904 an hour, which is exactly $3,600.00 across the 8,760 hours in a year.
Every comparison article treats Document AI as a per-page API and puts one number next to AWS and Azure. That number is right for four processors and wrong for eleven. Here is what Google's rate card actually meters.
Enterprise Document OCR, Form Parser, Custom extractor, Layout Parser, custom splitter and classifier
A 40-page PDF bills 40 pages.
Invoice parser, Expense parser, Utility parser
A 1-page invoice and a 10-page invoice cost exactly the same.
Bank statement, pay slip, W2, procurement and lending splitters
Page count does not appear in the price at all.
Custom processor hosting, provisioned service tier
You pay while it sits deployed and idle.
| Processor | Price | Unit Google bills in | Effective per 1,000 pages | 10-page rule applies |
|---|---|---|---|---|
| Invoice parser | $0.10 | per count, and 1 count is up to 10 pages | $10.00 to $100.00 | Yes |
| Expense parser (formerly receipt parser) | $0.10 | per count, and 1 count is up to 10 pages | $10.00 to $100.00 | Yes |
| Utility parser | $0.10 | per count, and 1 count is up to 10 pages | $10.00 to $100.00 | Yes |
| US driver license parser | $0.10 | per count | depends on the count definition | No |
| US passport parser | $0.10 | per count | depends on the count definition | No |
| Identity document proofing | $0.10 | per count | depends on the count definition | No |
| Bank statement parser | $0.75 | per classified document | not page-based | No |
| Pay slip parser | $0.30 | per classified document | not page-based | No |
| W2 parser | $0.30 | per classified document | not page-based | No |
| Procurement document splitter and classifier | $0.05 | per classified document | not page-based | No |
| Lending document splitter and classifier | $0.05 | per classified document | not page-based | No |
Google attaches the footnote "1 count equals up to 10 pages in a document" to exactly three rows: invoice, expense and utility parsers. The identity parsers are also priced per count but carry no such footnote, so we have not assumed the same rule applies to them. That distinction matters if you are budgeting driver license or passport work, and it is worth confirming with Google before you commit.
Invoice parser is $0.10 per count and a count is up to ten pages, so a count rounds up. Google publishes the examples, "between 1 to 10 pages costs $0.10" and "between 11 to 20 pages costs $0.20", but never converts them to the per-1,000-pages unit everyone actually compares on. Converted, the same processor at the same published rate spans a factor of ten.
| Document length | Counts billed | Cost per document | Effective per 1,000 pages | vs OCR at $1.50 |
|---|---|---|---|---|
| 1 page | 1 | $0.10 | $100.00 | 66.7x |
| 2 pages | 1 | $0.10 | $50.00 | 33.3x |
| 3 pages | 1 | $0.10 | $33.33 | 22.2x |
| 5 pages | 1 | $0.10 | $20.00 | 13.3x |
| 10 pages | 1 | $0.10 | $10.00 | 6.7x |
| 11 pages | 2 | $0.20 | $18.18 | 12.1x |
| 20 pages | 2 | $0.20 | $10.00 | 6.7x |
US invoices are overwhelmingly one or two pages. At that length invoice parser costs an effective $50.00 to $100.00 per 1,000 pages, which is above Form Parser's flat $30.00 and roughly 33 to 67 times the $1.50 you would pay to simply read the page. The crossover sits between three and four pages. A three-page invoice is $0.09 through Form Parser against $0.10 through invoice parser, so Form Parser wins; at four pages Form Parser reaches $0.12 and invoice parser is still $0.10, so it flips.
An eleven-page document buys a second full count. You pay $0.20 for eleven pages of value, so the effective rate jumps back up to $18.18 per 1,000 pages from $10.00 at ten pages. If your documents cluster just over a multiple of ten, splitting them before the call is a real and legitimate saving. Google also notes that synchronous requests do not accept documents over ten pages at all, so anything longer goes through the batch path anyway.
Graduated tiers applied, and the free first 1,000 pages of Enterprise Document OCR subtracted. The last column assumes one-page documents through invoice parser, which is the realistic case for accounts payable and also the worst case on this rate card.
| Monthly volume | Enterprise Document OCR | Layout Parser | Form Parser | Invoice parser, 1-page docs |
|---|---|---|---|---|
| 10,000 pages a month | $13.50 | $100 | $300 | $1,000 |
| 100,000 pages a month | $148.50 | $1,000 | $3,000 | $10,000 |
| 1,000,000 pages a month | $1,498.50 | $10,000 | $30,000 | $100,000 |
| 5,000,000 pages a month | $7,498.50 | $50,000 | $110,000 | $500,000 |
| 10,000,000 pages a month | $10,498.50 | $100,000 | $210,000 | $1,000,000 |
Prediction charges only. Custom extractors add $0.05 per hour per deployed version on top, and the provisioned service tier is separate again. Figures computed from the published rate card, 21 August 2026.
The long-form walkthrough of how each Document AI charge behaves, with worked examples.
Page counts, file sizes and rate limits, separately from cost.
The other Google OCR product, and when it is the cheaper call.
The Amazon side of the same decision, meter by meter.
The Microsoft side, where two products share the name OCR.
AWS, Azure, Google and the rest, normalized per 1,000 pages.
A rate card cannot tell you whether the fields come out right, and on Document AI it cannot even tell you the unit without reading a footnote. Upload one of your real documents, look at what comes back, and then decide whether you are buying an API call or a finished pipeline.
