Azure Document Intelligence Pricing: Form Recognizer Cost

Last updated August 2026. Azure Document Intelligence charges per page, not per seat. In the US East region on the standard S0 tier, optical character recognition costs $1.50 per 1,000 pages, the layout model and the prebuilt models (invoice, receipt, ID, tax forms) cost $10.00 per 1,000 pages, and custom extraction models cost $30.00 per 1,000 pages. Document classification is $3.00 per 1,000 pages and training a custom model is $3.00 an hour. There is a free tier, and monthly commitment tiers start at $190.

Every figure on this page came out of Microsoft's public retail prices API for the eastus region, queried on 29 August 2026. That matters because the Azure pricing calculator renders its rate table with JavaScript, so most articles about Azure Document Intelligence pricing quote a number somebody copied from another article two years ago. These are the live meter rates.

Azure Document Intelligence pricing table

Azure bills each capability on its own meter. You are not buying a plan, you are calling a specific model and paying for the pages that model reads. These are pay as you go S0 rates in US East.

Two things people miss reading this table. First, the meters stack. Running a classifier over a mixed batch and then the layout model costs $3.00 plus $10.00 per 1,000 pages, not $10.00. Second, a page is a page whether it holds one line or a dense table, and a 40 page bank statement is 40 billable pages.

How much does it cost to process 1,000 pages with Azure Document Intelligence?

Between $1.50 and $30.00, depending entirely on which model you call. OCR alone is $1.50. The layout model, which is the one that actually returns table structure, is $10.00. A custom model trained on your own form is $30.00. Most table extraction work lands at the $10.00 figure.

Put that against real volumes. A bookkeeping practice running 200 client invoices a month at two pages each is 400 pages, so about $4.00 a month on the prebuilt invoice model. An operations team pushing 10,000 pages of shipping documents through the layout model pays $100.00 a month. A lender processing 50,000 pages of statements on a custom model is looking at $1,500.00 a month before any add ons.

That spread is the whole argument for the per page model. At small volume it is close to free, and it stays proportional as you grow, which a per seat license never does.

What is the difference between Read, Layout, and the prebuilt models?

Read returns text and nothing else, which is why it is seven times cheaper. Layout returns text plus the table grid, selection marks and reading order, so it is the model you need if the output has to become a spreadsheet. Prebuilt models cost the same as layout but return named fields, such as invoice total or vendor name, rather than a raw grid.

The common and expensive mistake is calling Layout when Read would do, or calling a custom model when a prebuilt one already covers the document type. Microsoft ships prebuilt models for invoices, receipts, identity documents, US tax forms including W-2 and 1099, health insurance cards and bank statements. If your document is on that list, paying $30.00 per 1,000 pages to train your own is money burned.

Is Azure Document Intelligence free?

There is a free tier, and it is genuinely useful for evaluation rather than a trial that expires. It carries a monthly page allowance and rate limits, so it is sized for testing a model against your documents, not for running production volume. Once you exceed the allowance the S0 rates above apply, and Azure does not stop you at the boundary.

That last point deserves emphasis. Per page billing has no ceiling by default. A retry loop or a batch job pointed at the wrong folder bills every page it reads. Set a budget alert on the resource group before you put anything on a schedule.

What are Azure Document Intelligence commitment tiers, and are they worth it?

Commitment tiers are a fixed monthly fee for a page allowance, billed whether you use it or not, with a cheaper overage rate beyond the allowance. They only make sense once your volume is both large and predictable.

Read the smallest tier carefully before you sign up for it. Prebuilt 20K saves 5%, which is $10 a month, and you forfeit anything you do not use. If your volume swings between 8,000 and 25,000 pages a month, pay as you go is cheaper and simpler. The tiers earn their keep at the top of the table, where a million pages a month saves $2,500.

Azure also publishes connected container tiers at roughly 10 to 15% below the cloud rates, and disconnected container tiers billed annually for environments with no outbound internet. Those are procurement conversations, not self serve buttons.

How does Azure Document Intelligence pricing compare with AWS and Google?

All three cloud vendors bill per page, and the headline rates land close enough that price alone rarely decides it. Azure's layout model at $10.00 per 1,000 pages is the figure to compare, and both competitors charge separately for the table and forms features rather than bundling them, so the stacking behavior matters more than the sticker rate.

We worked through the same exercise for Amazon Textract pricing and its real cost per page, including how the table and forms charges compound on a single call. The pattern to watch across all three is identical: the cheap OCR meter is not the meter you will actually be billed on once you need structure.

Accuracy on your own documents should drive the choice, not a $2 difference per 1,000 pages. Run the same 50 page sample through each one and count the cells you have to fix by hand. That number swamps the rate difference at any volume a normal business runs.

Why is it called Form Recognizer in half the documentation?

Form Recognizer is the old name. Microsoft rebranded the service to Azure AI Document Intelligence, but the legacy meters still exist in the billing system and older SDK packages, tutorials and Stack Overflow answers still use the original name. If you see Form Recognizer in a code sample or on an invoice line, it is the same service.

The legacy Form Recognizer meters price prebuilt pages at $10.00 per 1,000 and custom pages at $50.00 per 1,000, which is higher than the current custom rate. Worth checking which meter your resource is actually provisioned against if you inherited it from someone else.

What are the hidden costs of Azure Document Intelligence?

The per page rate is honest, but it is not the whole bill. Four lines catch teams out.

• Stacked meters. Classifier plus layout plus an add on capability is three charges on the same page.
• Storage and egress. The documents live somewhere, usually Blob Storage, and the extracted JSON moves somewhere else.
• Engineering time. This is an API, not an application. Somebody writes the calling code, the retry logic, the error handling and the piece that turns JSON into a spreadsheet. That is the largest real cost for most teams and it never shows up on the Azure invoice.
• Retries on bad scans. Failed and re-run pages are billable pages.

The engineering line is the one to think hardest about. A cloud document API gives you structured JSON, and structured JSON is not a deliverable for the person who asked for the numbers. Somewhere downstream that output still has to be shaped and moved into the warehouse, ERP or accounting system where the rest of the business can see it, and that integration work usually outlasts the extraction work by a wide margin.

Is Azure Document Intelligence the right way to get tables out of PDFs?

It is the right answer for volume and automation, and the wrong answer for a person with a folder of statements. If you are processing thousands of pages a month inside an application you already run, $10.00 per 1,000 pages with no seat licenses is hard to beat, and the free tier lets you prove the accuracy before you commit.

If the actual situation is that someone in finance opens a PDF every week and retypes a table, an API is the wrong shape entirely. You would be hiring a developer to build a tool around a $4 a month bill. A desktop suite or a browser tool costs less than the first sprint. Our breakdown of the six ways PDF to Excel tools charge covers where each model stops making sense, and if you are weighing a desktop license instead, Foxit PDF Editor pricing is the cheapest credible suite in that category.

The bottom line on Azure Document Intelligence pricing

Budget $10.00 per 1,000 pages for real table extraction and treat everything else as a modifier. OCR only drops you to $1.50, custom models push you to $30.00, and stacked features add a few dollars per 1,000 on top. Commitment tiers are worth the paperwork above about 100,000 pages a month and not before.

Then add the engineering estimate, because that is the number that decides whether an API is the right tool. Compare the fully loaded figure against the converters that ship as finished products rather than against the raw per page rate, and the answer usually becomes obvious in either direction.

Prices in this article were read from the Microsoft retail prices API for the US East region on 29 August 2026 and are list rates before any enterprise agreement discount. Azure rates vary by region and change without much announcement, so confirm the current figure in the Azure pricing calculator before you build a budget on it.