Azure Document Intelligence Pricing 2026: Per-Page Rates, the Free Tier, Commitment Tiers and a Real Monthly Bill

How much does Azure Document Intelligence cost? The short answer

As of October 24, 2026, Azure Document Intelligence is pay per page with no subscription. Reading plain text (OCR) costs $1.50 per 1,000 pages. Every prebuilt model, including invoices, receipts, IDs, W-2s, contracts and layout, costs $10 per 1,000 pages. A custom extraction model costs $30 per 1,000 pages, and custom classification $3. The free tier covers 500 pages a month. A 25-client bookkeeping practice running about 5,000 pages a month through it pays roughly $70.

Microsoft now sells the service as Azure Document Intelligence in Foundry Tools; it used to be called Form Recognizer. Every price on this page comes from Microsoft's own Document Intelligence pricing page in US dollars, read October 24, 2026. Check the live page before you sign a budget.

Azure Document Intelligence pay-as-you-go prices

The paid tier is called S0. You are billed for each page the service analyzes, at a rate set by which model reads it.

Batch processing is priced the same as the regular calls for each model. The container version, for running it on your own servers, is priced the same as the cloud service.

Is Azure Document Intelligence free?

There is a free tier, called F0, with 500 pages a month across all models. Microsoft's footnote says it does not support the premium add-on features and does not include query fields. It is a way to test a model on your own documents, not a way to run a business's paperwork: a firm processing a few hundred invoices a month passes 500 pages in the first week or two.

A worked bill: a 25-client bookkeeping practice

Microsoft's page lists rates, not workloads, so here is one. A bookkeeping practice handles AP for 25 small-business clients. Each month clients forward about 2,400 vendor invoices averaging 1.5 pages, 1,200 single-page receipts, and 30 bank statements of 8 pages that the prebuilt models do not cover. Everything lands in one shared inbox, so the system sorts it first.

That is under 3 cents a client document. The extraction itself is rarely the expensive part of an invoice pipeline; the hours go into checking low-confidence fields and getting the data into the accounting system, which is why our invoice processing automation guide spends most of its time on matching and approvals.

The page-count trap that inflates the bill

Microsoft's FAQ is specific: when a document is submitted, all of its pages are analyzed unless the request names a page range. Vendors love attaching their terms and conditions. If 1 invoice in 5 in the example above carries 4 extra pages of fine print, that is 480 invoices x 4 pages = 1,920 more pages at $10 per 1,000, or $19.20 a month. That is a 27% higher bill for pages nobody wanted read.

The fix costs nothing: have the automation send only the first page or two of each invoice to the prebuilt model, and send the rest to Read at $1.50 per 1,000 if you need it searchable. Two other habits keep the bill honest. Run the $3 classifier before the expensive models, so a statement never goes through the invoice model by mistake. And keep custom extraction, at three times the prebuilt price, for documents no prebuilt model covers.

When commitment tiers pay off

Azure also sells monthly commitment tiers: you prepay for a block of pages at a discount and pay a set rate per 1,000 pages over it.

The smallest prebuilt tier saves only 50 cents per 1,000 pages, so it breaks even at 19,000 prebuilt pages a month. Below that, pay as you go is cheaper. The practice above sends 4,800 pages through prebuilt models and would pay $190 on the tier instead of $48. A business needs roughly four times that volume before a commitment makes sense. Microsoft also sells yearly commitments for disconnected containers, starting at $8,640 a year for prebuilt, which is for sites that cannot send documents to the cloud at all.

Azure Document Intelligence vs AWS Textract on the same work

Textract prices below are from Amazon's Textract pricing page, read October 24, 2026, using its US West (Oregon) examples. Our AWS Textract pricing breakdown runs a different workload through every Textract feature.

On invoices and receipts the two list prices tie exactly. Run the same practice through Textract and it costs about $48 for invoices and receipts plus $4.80 for the statements (Tables and Queries at $0.020 a page), about $53, but with no sorting step priced in; the $15 Azure spends on classification is the difference. Pick the cloud your other systems already run on, because the extraction bill is too small to decide it.

What Microsoft's price page leaves out

• Storage and the rest of the pipeline. Documents usually sit in Azure Blob Storage, and something has to call the service, check the results and write them out. None of that is on this page.
• Review time. The models return fields with confidence scores. Somebody, or a rule, has to decide what happens to the low-confidence ones.
• Separate bills by model. Microsoft says it cannot split one bill by model, though the Azure portal's metrics tab shows usage by model type.
• Which fields come back. The prebuilt price is the same for every prebuilt model, but each returns a fixed schema. If a field you need is missing, it is a query field ($10 per 1,000 more) or a custom model ($30).

Is Azure Document Intelligence worth it?

For a small business already on Microsoft 365 or Azure, yes, as the extraction step inside a larger system. At $10 per 1,000 pages it is far cheaper than a person keying invoices, and cheaper than most per-document tools in our Docsumo vs Rossum vs Nanonets comparison, which bundle review screens and integrations into the price. What you give up is that bundle: Azure gives you an API, not an AP workflow.

So the honest comparison is not Azure against a person. It is Azure plus a build against an off-the-shelf AP tool. If your invoices then get paid through an AP platform anyway, read our Bill.com review first and check what the AP tool already captures before you pay to build the same step.

Who can build a document processing workflow on Azure for my business?

A developer comfortable with Azure can wire up one document type in a few days. A builder earns a fee when the system has to sort a mixed inbox, apply your approval rules, handle low-confidence fields and write into QuickBooks, Xero or your ERP. SuperDupr, the AI ops shop Justin McKelvey runs, builds those pipelines. Our published price for AI document processing is $8,000 to $20,000. Most AI builds go live in 2 to 4 weeks, support afterwards runs $200 to $2,000 a month depending on volume, and the business owns what we build. Azure usage stays billed to your own Azure account at Microsoft's rates; we do not resell it. The first step is a free 30-minute strategy session, booked from that page.

Sources

• Azure Document Intelligence in Foundry Tools pricing, azure.microsoft.com/en-us/pricing/details/ai-document-intelligence, US dollars, read October 24, 2026: free tier (500 pages a month, footnote on premium features and query fields), S0 pay-as-you-go rates for Read, prebuilt models, custom classification, custom extraction, custom generative extraction, add-ons, query fields, training and batch; commitment tiers for Azure, connected and disconnected containers; FAQ on page counting, the free tier, no subscription, training charges, per-model billing and container pricing.
• Amazon Textract pricing, aws.amazon.com/textract/pricing, read October 24, 2026: free tier and the per-page rates in its US West (Oregon) pricing examples.
• Worked bill, page-count example and break-even: SuperDupr's arithmetic on the published rates above, for an assumed 25-client practice.