Azure Document Intelligence Pricing 2026: Official Rates Per 1,000 Pages

Last updated July 25, 2026. Every rate below was re-verified on that date against Microsoft's official Azure Retail Prices API (product "Azure Document Intelligence", region East US), which is the same price feed the Azure portal bills from.

Azure AI Document Intelligence is priced pay-as-you-go per page on the S0 tier. The official 2026 rates are $1.50 per 1,000 pages for the Read (OCR) model, $10 per 1,000 pages for Layout, $10 per 1,000 pages for prebuilt models such as invoice and receipt, $30 per 1,000 pages for custom extraction, $3 per 1,000 pages for the document classifier, $10 per 1,000 pages for query fields, and $6 per 1,000 pages for each add-on capability. Read drops to $0.60 per 1,000 pages above 1 million pages a month, and custom extraction drops to $20. A free F0 tier covers 500 pages a month but returns only the first two pages of any request. Below the pay-as-you-go table there is a second, much less visible price list: commitment tiers that take Read as low as $0.45 per 1,000 pages and custom extraction to $18. Those rates do not appear on Microsoft's marketing pricing page at all.

Azure Document Intelligence pricing per 1,000 pages (official 2026 rates)

Azure bills by the page and by the model you call. There is no platform fee, no seat license, and no minimum spend on pay-as-you-go. These are the S0 standard-tier rates, read directly from Microsoft's retail price feed on July 12, 2026 (East US; a few regions differ by a small amount).

One document can hit more than one meter. A non-standard form typically runs through the classifier to route it ($3 per 1,000) and then custom extraction to read it ($30 per 1,000), so the real per-document cost is the sum of the models that fire, not a single rate. That is the single most common mistake in an Azure estimate.

What is the official Azure Document Intelligence Read (OCR) price per 1,000 pages?

The official Read model price is $1.50 per 1,000 pages on the S0 tier, falling to $0.60 per 1,000 pages for volume above 1 million pages in a month. Read is the plain OCR model: it returns text, lines, words, and handwriting, with no key-value pairs and no tables. At $1.50 it matches AWS Textract Detect Document Text and Google's Enterprise Document OCR to the cent, so plain OCR is a commodity across all three clouds and price is not a reason to choose between them.

Azure Document Intelligence commitment tier pricing (the rates Microsoft does not put on the pricing page)

If your volume is steady, Azure sells prepaid commitment tiers that are materially cheaper than pay-as-you-go. They are hard to find: the marketing pricing page shows the pay-as-you-go rates, and the commitment tier rates are exposed through the Azure Retail Prices API and the portal's commitment tier blade. Each tier is a fixed monthly fee for an included page allowance, and pages beyond the allowance bill at that same discounted rate, so the effective price per 1,000 pages is the same whether you land under or over the commitment.

Two things stand out. First, the commitment discount is steep on Read (up to 70% off) and shallow on prebuilt (25% at a million pages a month), so committing helps a high-volume plain-OCR pipeline far more than an invoice pipeline. Second, the entry Read commitment starts at 500,000 pages a month, which is well above what most teams process, so for anything under roughly half a million pages a month the commitment tiers are not available to you and pay-as-you-go is the only path.

There is also a parallel set of rates for connected containers, which let you run Document Intelligence on your own infrastructure while still metering to Azure. They run below the cloud commitment rates: the 1,000,000-page prebuilt tier is $6,000 a month ($6.00 per 1,000 pages, 20% below the equivalent cloud tier), the 1,000,000-page custom tier is $15,300 ($15.30 per 1,000, 15% below), and the 16,000,000-page Read tier is $6,480 ($0.41 per 1,000, 10% below). If you have a data-residency or air-gap requirement, the container path is not a price penalty. It is a small discount.

Why does my Azure bill say Form Recognizer instead of Document Intelligence?

Because Microsoft never retired the old billing product. Azure Form Recognizer was renamed to Azure AI Document Intelligence in July 2023, but the Azure Retail Prices API still returns two separate live products, both now under the service name Foundry Tools: Azure Document Intelligence with 68 meters, and a legacy Form Recognizer product with four. They do not price custom extraction the same way. The legacy product's S0 Custom Pages meter bills $50 per 1,000 pages, effective since August 2021, against $30 per 1,000 under the current name, effective June 2024. We verified that in East US, West US 2 and North Europe on August 10, 2026.

That does not prove which meter your own subscription is billed on, because it depends how the resource was provisioned. It does mean the product heading on your invoice is not enough to go on: find the line item and read the meter name. At 100,000 custom pages a month the gap between the two meters is $2,000. The full breakdown, including every commitment tier and the three names Microsoft now uses for one service, is on our Azure Form Recognizer pricing reference, and the API and SDK changes behind the rename are covered in Azure Form Recognizer vs Document Intelligence.

Does batch analysis cost less on Azure Document Intelligence?

No. Azure meters batch analysis at exactly the same rate as a synchronous call: batch Read is $1.50 per 1,000 pages, batch Layout is $10, batch prebuilt is $10, batch custom extraction is $30, and batch add-ons are $6. There is no batch discount. This surprises teams coming from Mistral or Gemini, both of which cut batch or asynchronous inference by 50%. On Azure, batch is a throughput and convenience feature, not a cost lever. The only real cost levers Azure gives you are picking a cheaper model and committing to volume.

Is there a free tier for Azure Document Intelligence?

Yes. The free F0 tier processes up to 500 pages a month at no cost and covers every model, but it returns only the first two pages of each request and caps throughput at 1 transaction per second. That two-page truncation catches most evaluators off guard: send a 10-page contract to F0 and you get results for pages 1 and 2, which reads like a broken model rather than a tier limit. F0 is fine for a proof of concept on single-page documents and useless for production. Custom model training also includes 10 free hours a month on the paid tier, after which training compute bills at $3 an hour.

What is the official Azure custom extraction price per 1,000 pages?

Custom extraction is $30 per 1,000 pages on S0 with no pay-as-you-go volume tier at all, which surprises people who assume it discounts like Read does. Microsoft's price feed returns a single custom-pages rate with a tier minimum of zero, so the only way below $30 is a commitment tier, which runs from $27 per 1,000 at 20K pages down to $18 per 1,000 on the largest. It is the most expensive standard meter Azure runs, ten times its own prebuilt rate and twenty times Read. Custom generative extraction sits at the same $30 but does not get the volume step-down. The practical rule: if a prebuilt model covers your document type, use it. Moving 100,000 invoice pages a month from custom extraction to the prebuilt invoice model takes the bill from $3,000 to $1,000.

How much does Azure Document Intelligence cost at real volumes?

Worked examples at the official July 2026 rates, API charges only.

The archive row is worth staring at. At 250,000 pages a month you pay $375 pay-as-you-go. The 500,000-page Read commitment also costs $375. If you are anywhere near half a million pages, the commitment tier is free money, and Microsoft does not surface it where you would look.

How is Azure Document Intelligence priced?

Azure Document Intelligence is priced per page analyzed, billed monthly, with the rate set by the model you call rather than by a plan you pick. A page is one side of a document, so a 10-page PDF through Layout counts as 10 pages. Word and HTML files are counted in 3,000-character blocks per page, each Excel worksheet counts as one page, and each PowerPoint slide counts as one page. There is no subscription and no minimum spend on pay-as-you-go, so a month with zero traffic costs zero.

Does Azure Document Intelligence charge for failed or low-confidence reads?

Azure bills for every page the service analyzes, whether or not the extracted values are right, so a low-confidence or partly wrong read still counts as a billed page. Accuracy does not change the API charge. It changes your downstream cost, because every uncertain field becomes manual correction time. A low per-page rate with a high correction rate can easily cost more in total than a higher rate that needs almost no cleanup, which is why review time belongs in any honest estimate.

What does the Azure per-page price not include?

The per-page rate covers the API call, not the working system around it. Azure Document Intelligence is a cloud service a developer assembles into a pipeline, so real production cost also includes the engineering to call the API, the classification routing you wire up, a human review screen for low-confidence reads, validation rules, the export step into your accounting or ERP system, and the Azure subscription and storage the workload sits on. For non-standard layouts you also spend time labeling documents and training custom models in Document Intelligence Studio. None of that appears on the pricing page, and for many teams it is the larger number.

How do you estimate your Azure Document Intelligence bill?

Estimate in four steps. Count the pages you process a month. Decide which model each document type needs. Multiply pages by that model's rate and add any add-ons or query fields. Then check whether your volume clears a commitment tier, because the step-down is large on Read and real on custom. For example, 50,000 invoice pages on the prebuilt invoice model is $500 a month in API charges; the same volume on custom extraction is $1,500. After that, add the costs the per-page rate hides, the build and the review time, to reach a number you can defend in a budget meeting.

Is Azure Content Understanding cheaper than Azure Document Intelligence?

Yes, on every content extraction meter, and if you are budgeting an Azure document pipeline in 2026 you need to know this before you sign anything. Microsoft now sells a second document AI service, Azure AI Content Understanding, which reached general availability on API version 2025-11-01. It undercuts Document Intelligence on the identical job: $1.00 per 1,000 pages for OCR against Document Intelligence's $1.50, and $5.00 per 1,000 pages for layout against $10.00. Digital files are the extreme case. Content Understanding bills DOCX, XLSX, HTML and email at a Minimal meter of $0.01 per 1,000 pages because no OCR is performed, while Document Intelligence has no such meter and charges the full Read rate of $1.50. On a pipeline of Word and Excel files that is a 150x difference.

All rates per 1,000 pages, East US, read from Microsoft's Azure Retail Prices API on July 13, 2026.

Two things stop this from being a straight win. First, Document Intelligence is not deprecated: Microsoft has announced no end of support and every meter above is still live, so anyone telling you it is being killed off is guessing. Second, Content Understanding has no flat rate for structured field extraction. It charges $5.00 per 1,000 pages of content extraction plus $1.00 of contextualization, and then the tokens of a Foundry model you deploy and pay for yourself. Microsoft's own worked example for invoices comes to $8.37 per 1,000 pages on a GPT-4.1-mini deployment, but its own note says a full GPT-4.1 deployment lifts the same job to roughly $33, which is worse than the $30 flat rate you already have here. The model you attach, not the service you pick, decides whether you save money.

Document Intelligence still wins outright on three counts: a predictable flat $30 per 1,000 pages for custom extraction, a $3 per 1,000 pages classifier that Content Understanding has no cheap equivalent for, and a 2,000-page-per-call ceiling against Content Understanding's 300. The full meter-by-meter breakdown is on our Azure Content Understanding pricing reference, and the capability and migration decision is laid out on Content Understanding vs Document Intelligence.

Azure Document Intelligence vs a ready-to-use extraction product on cost

A raw cloud service wins on headline per-page price because you supply everything around it. A ready-to-use document extraction product costs more per page on paper but bundles the classification, review, validation, and export that you would otherwise build and maintain. The honest comparison is not rate against rate; it is the Azure per-page rate plus your build and operating time against the product's all-in per-page price. DocuOCR is a ready-to-use product: it classifies a mixed file, extracts the fields you define, validates them, routes low-confidence reads to review, and exports clean data through a dashboard and one REST API, with no Azure subscription, no pipeline to assemble, and no custom-model training project. If your team has the engineers and the volume to run a cloud service, Azure can be cheaper per page. If you want the workflow already built, a product is usually cheaper once you count the build.

How Azure Document Intelligence pricing compares to AWS and Google

Azure is not priced in a vacuum, and on several lines it is the cheapest of the three major clouds. All three charge $1.50 per 1,000 pages for plain OCR, verified July 2026. For structured extraction, Azure custom extraction and Google's Custom Extractor both sit at $30 per 1,000 pages, while AWS Textract charges $50 per 1,000 for Forms and $70 for Forms, Tables, and Queries together. Azure is also the cheapest place to classify a document, at $3 per 1,000 pages against Google's $5, and unlike Google it charges no idle hosting fee for a deployed custom model (Google bills $0.05 an hour per deployed processor version, roughly $438 a year, whether or not you send it traffic).

See the full side-by-side in our OCR API pricing comparison, the normalized OCR pricing per 1,000 pages reference that puts every vendor on the same unit, or the single-vendor guides for AWS Textract pricing and Google Document AI pricing. If your volume is large enough that the commitment tiers matter, our breakdown of what it costs to OCR 1 million pages runs the math across every vendor.

The $1.50 row is only equal at list price. AWS and Azure both drop that rate to $0.60 per 1,000 pages above one million pages a month, while Google Enterprise Document OCR does not reach its $0.60 tier until above five million. So at two million pages a month Azure pay-as-you-go and AWS both bill about $2,100 and Google bills about $3,000, and an Azure 2M commitment tier brings the same volume down to $1,200. Azure ends up the cheapest of the three at every volume above half a million pages, which is not visible on any list price comparison. The full curve is on high volume OCR API pricing.

One more axis is worth checking before you pick on price alone. The three clouds give three different default answers to what happens to a document after the extraction returns: Azure deletes it 24 hours after the analyze request and keeps it in the same region, Google states it never trains on customer data, and AWS may store content for service improvement unless you set an Organizations opt-out policy that ships switched off. We put all three side by side, with the quotes, in the OCR API data retention comparison.

Where the Azure price sits in a real workflow

The model you pick drives most of the cost, but the document type drives the model. Teams pushing invoice data into accounting usually pair extraction with accounts payable automation software so approved invoices flow straight into the AP workflow rather than stopping at a JSON file. If you are still deciding which Azure model your documents need, our Azure Document Intelligence limits page covers the file size and page ceilings each model enforces, and Azure Document Intelligence alternatives compares the workflow, not just the rate.

Does handwriting cost extra on Azure Document Intelligence?

No. Handwriting is read by the Read and Layout models at their normal rates, with no handwriting meter and no add-on to enable. What changes is language coverage, and it changes by API version. The v4.0 GA Read model supports 12 handwritten languages against more than 300 for printed text. The v3.0 and v3.1 models support nine of those, without Russian, Thai or Arabic, and Microsoft states plainly that "Document Intelligence v2.1 does not support handwritten text extraction". The full side-by-side is in our handwriting OCR reference.