Google Document AI Pricing 2026: Official Rates Per Page

Last updated July 25, 2026. Every rate below was re-verified on that date against the official Google Document AI pricing page.

Google Document AI pricing is pay-as-you-go by the page, with a different rate for each processor. The published rates run $1.50 per 1,000 pages for the Enterprise Document OCR processor, $10 per 1,000 pages for the Layout Parser, $30 per 1,000 pages for the Form Parser and the Custom Extractor, $5 per 1,000 pages for a Custom Classifier or Splitter, and $10 per 1,000 pages for the prebuilt Invoice and Expense parsers. On top of the per-page rate, every deployed custom processor version costs $0.05 per hour to host, which is about $438 a year per version whether you send it a page or not. This guide breaks down the 2026 rates, the volume thresholds, the hosting fee most estimates miss, and what the per-page number does not cover.

Google Document AI pricing per 1,000 pages (2026)

Document AI is a family of processors rather than a single service, and you pay for the processor you call. Here are the published rates, with the discounted tier that applies once monthly volume crosses the threshold shown.

Note the shape of that table. Plain OCR is cheap and structured extraction is 20x more expensive, which is the same pattern every major cloud vendor uses. The Layout Parser sits in the middle and is often the right call for teams feeding documents to a language model rather than to a database.

What are the Google Document AI processor types?

The processor type is an uppercase identifier, and it is not the name you see in the console. You select "Enterprise Document OCR" in the Cloud console, but the string the API expects is OCR_PROCESSOR. "Form Parser" is FORM_PARSER_PROCESSOR, "Layout Parser" is LAYOUT_PARSER_PROCESSOR and "Invoice Parser" is INVOICE_PROCESSOR. Searching Google's documentation for the friendly display name will not return the constant your code needs, which is why so much copied sample code fails on the first call.

The document-specific parsers follow the same convention and bill at $10 per 1,000 pages: INVOICE_PROCESSOR, EXPENSE_PROCESSOR, BANK_STATEMENT_PROCESSOR, FORM_W2_PROCESSOR, PAYSTUB_PROCESSOR, US_DRIVER_LICENSE_PROCESSOR and ID_PROOFING_PROCESSOR. The equivalent identifiers for AWS Textract and Azure, along with what each one returns, are mapped side by side on OCR API operations and model IDs.

How much do the prebuilt Document AI processors cost?

Google prices some prebuilt processors per page and others per document, which trips up a lot of budgets. A one page pay slip and a nine page pay slip cost the same under per-document pricing. A twelve page invoice does not.

The invoice rounding rule is worth reading twice. Billing is in blocks of ten pages, so an eleven page invoice bills as twenty pages, at $0.20. If your invoices routinely run just over a ten page boundary, your real invoice rate is closer to $20 per 1,000 pages than $10. Some processors, including the Utility and Procurement parsers, are limited access and require you to request approval before you can use them at all.

What is the hidden cost in Google Document AI pricing?

Processor hosting. Every deployed version of a custom processor costs $0.05 per hour, which works out to roughly $438 per version per year, and it accrues whether or not you send it a single page. A team that trains three custom extractors and keeps two versions of each deployed for rollback is paying about $2,600 a year before processing anything.

That fee is the single most common surprise on a Document AI bill, because it does not scale with usage and therefore never appears in a per-page comparison. Two habits keep it under control: undeploy old processor versions once a new one is stable, and resist the temptation to keep a processor per document type when one well trained extractor covers several.

Provisioned capacity is the other line item. If you need guaranteed throughput rather than best effort, Google charges about $300 per month for each additional page-per-minute of provisioned capacity. Most teams do not need it. The ones that do tend to discover it during a month-end batch run.

Does Google Document AI have a free tier?

Document AI does not publish a standing monthly free allowance the way some vendors do. New Google Cloud accounts get trial credit through the Google Cloud Free Program, and you can spend that credit on Document AI, but once it is gone every page bills at the standard rate. Plan on paying for a real evaluation.

There is one genuinely useful billing behavior: failed requests that return a 4xx or 5xx response are not billed. A malformed PDF that Document AI rejects does not cost you anything. A malformed PDF that Document AI accepts and reads badly does.

How much does Google Document AI cost at volume?

Work from your own page mix. Three examples at 50,000 pages a month:

• Plain OCR: 50,000 pages of Enterprise Document OCR at $1.50 per 1,000 is $75 a month.
• Invoices through the prebuilt parser: 50,000 pages at $10 per 1,000 is $500 a month, before the ten page rounding effect.
• Custom fields through a Custom Extractor: 50,000 pages at $30 per 1,000 is $1,500 a month, plus about $37 a month in processor hosting per deployed version.

The discounted tiers are far away. OCR only gets cheaper above five million pages a month, and the Form Parser and Custom Extractor only get cheaper above one million. A US mid-market team running 50,000 or 300,000 pages a month pays the standard rate on every single page, so build the budget on the standard rate and treat the volume discount as something that will never happen.

That five million page boundary is worth comparing against the other two clouds, because it is where Google quietly becomes the expensive option. AWS Textract and Azure Read both begin their $0.60 tier above one million pages a month. Google does not begin its $0.60 tier until above five million. All three list at the same $1.50, so a company running two million pages a month pays about $2,100 on AWS and about $3,000 on Google for broadly comparable basic OCR, purely because of where the boundary sits.

Google's structured meters do not behave this way: Form Parser and Custom Extractor both tier at 1,000,001 pages, the same boundary the other two use. It is specifically the basic OCR meter where the discount arrives five times later. The full curve across every vendor and meter is on high volume OCR API pricing.

What does the Google Document AI per-page price not include?

The per-page rate buys a JSON response from an API inside a Google Cloud project. It does not buy a working process. Budget separately for:

• Engineering. Setting up the GCP project, service accounts and IAM, the Cloud Storage buckets, the batch or online request flow, and the code that maps Document AI entities to your own field names.
• Training data. A Custom Extractor is only as good as the documents you label. Someone has to gather and annotate them, and re-label when a vendor changes their layout.
• Human review. Document AI returns confidence scores. The screen where a person fixes a 68% confident total, and the audit trail showing who fixed it, is software you write. Google offers a separate human-in-the-loop capability that bills on its own.
• Surrounding GCP services. Cloud Storage, Cloud Functions or Cloud Run, Pub/Sub, logging, and egress all land on the same invoice.
• Validation and export. Confirming that line items sum to the total, that the invoice date is plausible, and that the result reaches your ERP is application code.

Google Document AI vs a ready-to-use extraction product on cost

The honest comparison depends entirely on what you are trying to do. If you have a Google Cloud team and you need raw text at scale, Enterprise Document OCR at $1.50 per 1,000 pages is excellent value and you should use it.

If you need structured, validated fields, the arithmetic shifts. A Custom Extractor is $30 per 1,000 pages plus hosting plus labeling plus the pipeline plus the review screen. DocuOCR's published plans work out to roughly $14 to $20 per 1,000 pages depending on volume, and that includes classification, extraction, validation, the human review screen, the dashboard, and export, with no GCP project, no service account, and no processor versions to deploy or undeploy.

Neither answer is universally right. A platform engineering team already standardized on Google Cloud, processing millions of pages, will do well on Document AI. A finance or operations team that needs clean data out of 20,000 mixed pages this month, and does not have an engineer to spare, is comparing $30 per 1,000 pages plus a project against $15 per 1,000 pages plus an afternoon.

How do you estimate your Google Document AI bill?

Count pages from a real month, not files. Split them by the processor each page needs: OCR only, Layout, Form Parser, Custom Extractor, or a prebuilt parser. Multiply each bucket by its rate. Then add the hosting fee for every processor version you intend to leave deployed, add the rounding penalty if your invoices cross ten page boundaries, add 10 to 20 percent for Cloud Storage and compute around the API, and add the engineer who will own the pipeline.

Watch out for per-document processors specifically. At $0.75 per document, the Bank Statement Parser is cheap for a 40 page statement and expensive for a one page summary. If your end goal is simply to turn a PDF statement into a clean spreadsheet, price that path directly instead of assuming a raw API is the shortest route to it.

The full official Google Document AI rate card (July 2026)

Two lines on that table are the ones that surprise people, and both are covered in detail below: the 10-page block on the prebuilt parsers, and the hosting fee that runs whether or not you send a page.

Does Google Document AI have a batch discount?

No. Google's batch endpoint exists to lift the limits, not the price. Online requests cap at 15 pages and 40 MB, which is what raises PAGE_LIMIT_EXCEEDED; batch takes 1 GB a file, up to 1,000 pages a file on Enterprise Document OCR, and 5,000 files a batch, at exactly the same per-page rate. Azure and AWS do not discount batch either. Only Mistral and Gemini do, both by 50%.

Does Google Document AI charge for failed requests?

No, and this is a genuine and rare advantage. Google does not bill requests that return a 4xx or 5xx error. Azure and AWS bill for every page the service analyzes regardless of the outcome. It matters more than it sounds: a pipeline that hits rate limits, malformed files or oversized PDFs at any real rate will retry, and on Google those failures are free while on Azure and AWS they are billed pages. What Google does still charge for is a successful request that returns a wrong answer, because no vendor refunds a bad read.

How much does the Google Document AI Invoice Parser cost per page?

The Invoice Parser is billed at $0.10 for every 10 pages in a document, and the block is the billing unit, not the page. That means a one-page invoice and a ten-page invoice cost exactly the same: $0.10. Because most business invoices are one or two pages, the effective rate people actually pay is close to $100 per 1,000 invoices, not the $10 per 1,000 pages the block rate appears to promise.

That is a ten-fold difference from the two obvious rivals, and it is worth seeing side by side.

And no, you cannot dodge it by merging ten invoices into one PDF. The parser extracts one invoice per document, so a merged file returns a single blended record rather than ten clean ones. The block discount only rewards genuinely long documents.

The useful consequence is on the last row. If invoices are the bulk of your volume, Google's own Custom Extractor at a flat $30 per 1,000 pages is cheaper than Google's prebuilt Invoice Parser on single-page invoices, by a factor of more than three. You trade the prebuilt schema for training your own, which is real work, but the arithmetic is hard to argue with. We work through the whole thing, including the page-count math for your own invoice mix, on Google Document AI Invoice Parser pricing.

The short answer on Google Document AI pricing

Enterprise Document OCR is about $1.50 per 1,000 pages, the Layout Parser is $10, the Form Parser and Custom Extractor are $30, classifiers and splitters are $5, and prebuilt parsers like Invoice and Expense work out to about $10 per 1,000 pages with ten page rounding. Volume discounts start at one million pages for extraction and five million for OCR, so most teams pay standard rates. The line that catches people is processor hosting at $0.05 an hour, roughly $438 a year for every custom processor version left deployed. Failed requests are not billed, but badly read pages are.

For the full cross-vendor picture, see our OCR API pricing comparison, which puts Google, Azure, and AWS rates side by side. For the single-vendor breakdowns, read our Azure Document Intelligence pricing guide and our AWS Textract pricing guide. If you would rather not run a GCP project to read a document, DocuOCR is a ready-to-use Google Document AI alternative that returns validated fields instead of raw entities.

For every vendor's rate on one normalized unit, including the per-document and per-token vendors, see our OCR pricing per 1,000 pages reference. The 10-page block is one of several ways the billing unit beats the rate, covered in per page vs per document vs per token. At enterprise volume, see what it costs to OCR 1 million pages, and for the ceilings that force you onto batch, Google Document AI limits.

Does handwriting cost extra on Google Document AI?

No. Handwriting is a default detection attribute of Enterprise Document OCR and bills at the standard $1.50 per 1,000 pages. None of the three premium OCR add-ons, billed at $6 per 1,000 pages, is handwriting: they are Math OCR, checkbox extraction and font style detection. Unlike AWS and Microsoft, Google publishes no handwritten-language table anywhere in its documentation, so coverage for a given language has to be tested rather than looked up. See our handwriting OCR reference for the comparison.