Azure AI Content Understanding Pricing: Cost per 1,000 Pages and How It Compares With Document Intelligence

Azure AI Content Understanding pricing, every meter

Microsoft does not publish these numbers in a readable table. The Azure pricing page for Content Understanding renders its rates as placeholders, the same way the Document Intelligence page has for most of 2026. Every figure below was read directly from the Azure Retail Prices API on 21 September 2026, filtered to productName eq 'Azure Content Understanding'. Nothing here is estimated.

Content extraction: available in 16 regions, including every major US region

Content extraction is the OCR half of the service. It reads a file and returns text, layout and tables. It does not return named invoice fields. All rates are USD per 1,000 pages.

Two token meters ride alongside those page rates when the service has to reason over what it read: standard contextualization at $0.001 per 1,000 tokens and advanced contextualization at $0.003 per 1,000 tokens. Audio content extraction is $0.36 an hour and video content extraction is $1.00 an hour.

The synchronous surcharge is exactly 50 percent

Basic extraction is $1.00 asynchronously and $1.50 synchronously. Standard is $5.00 asynchronously and $7.50 synchronously. The output is the same; you are paying purely for not having to poll for a result. That is worth knowing because Azure Document Intelligence prices its batch and sync meters identically, so a team migrating from Document Intelligence and keeping a synchronous call pattern will see a 50 percent line-item increase that has nothing to do with the work performed.

Field extraction: the meter that matters for invoices, and the one with a catch

Field extraction is the half that returns typed values: invoice number, vendor, due date, totals, line items. This is what an accounts payable team is actually buying. It is priced separately, and it is not sold everywhere.

Field extraction also carries its own token meters, and here the regional arithmetic flips. Standard field extraction input is $0.00275 per 1,000 tokens in West US and $0.003025 in Sweden Central; output is $0.011 against $0.0121. Pro field extraction input is $0.00121 against $0.00133, output $0.00484 against $0.00532. Classification input is $0.00344 against $0.00378 and classification output $0.01375 against $0.015.

So Sweden Central is 10 percent cheaper on every per-page and per-hour meter, and 10 percent more expensive on every per-token meter. Which region is cheaper for you depends on whether your workload is dominated by pages or by the reasoning tokens spent on them.

Field extraction runs in three regions worldwide, and only one of them is in the United States

This is the single most important thing on this page and it is not stated on any Microsoft pricing page. Content extraction is offered in 16 regions, including East US, East US 2, Central US, North Central US, South Central US, West US, West US 2 and West US 3. Field extraction is offered in exactly three: West US, Sweden Central and Australia East.

If your Azure footprint is in East US, which is the default region in most quickstarts and the home region for a large share of US tenants, you cannot run Content Understanding field extraction there at all. You can OCR a document in East US. You cannot ask the service to return an invoice number. Getting typed invoice fields means provisioning in West US and accepting cross-region data movement, which is a conversation with your security team before it is a conversation about price.

Content Understanding against Azure Document Intelligence, meter by meter

Microsoft has folded Document Intelligence into Content Understanding in Foundry Tools, so the natural question is whether the newer service is cheaper. On raw reading it is. On invoice field extraction it is emphatically not.

Read that fourth row carefully, because it inverts the usual assumption that the newer Azure service is the cheaper one. For the specific job of turning a vendor bill into structured data, the prebuilt invoice model on Document Intelligence costs $10.00 per 1,000 pages and Content Understanding costs $22.00. On 10,000 invoices a month that is $120 against $220, and on 100,000 it is $1,000 against $2,200.

The reason is architectural rather than commercial. The prebuilt invoice model is a fixed schema that Microsoft trained once and serves cheaply. Content Understanding field extraction runs a generative model against a schema you define, which is why it can read documents the prebuilt model has never seen, and why it costs more than twice as much when the prebuilt model would have done the job.

Content Understanding has no commitment tiers, no volume discount and no disconnected option

Every Content Understanding meter in the Retail Prices API carries the pricing type Consumption. There is not a single commitment tier, reserved capacity SKU or disconnected container product in the catalog. We checked for all three.

Document Intelligence has all of them. Its prebuilt commitment tiers start at $190 a month for 20,000 pages, an effective $9.50 per 1,000, and run to $7,500 a month for 1,000,000 pages at $7.50 per 1,000. Connected containers are 20 percent cheaper again, so a 20,000 page container commitment is $152 a month, an effective $7.60 per 1,000. Disconnected, air-gapped licences exist at $8,640 a year for 100,000 pages a month, which annualizes to exactly the connected price.

That gap matters more than the headline rates for anyone at volume. A finance team running 100,000 invoice pages a month pays $2,200 on Content Understanding field extraction with no discount available at any volume, against $900 a month on a Document Intelligence prebuilt commitment tier. That is a 2.4 times difference, and it grows rather than shrinks as you scale.

Where each service is the right answer

Content Understanding is the better buy when you are paying for OCR or layout at volume, when a meaningful share of your files are digital Word, Excel, HTML or email documents that hit the $0.01 meter, when your documents vary so much that no prebuilt schema fits, or when you also need audio and video in the same pipeline.

Document Intelligence remains the better buy when you want a flat, forecastable per-page rate, when the prebuilt invoice or receipt model already returns the fields you need, when you need a commitment tier or an air-gapped deployment, and when your Azure footprint is not in West US.

For most US accounts payable teams reading this page, that means the older service is still the cheaper one, and the migration Microsoft is nudging you toward will raise your invoice extraction bill unless you are also consolidating a lot of OCR spend.

What neither Azure service gives you

Both meters buy you an API, not a working accounts payable process. Whichever one you pick, you are still building the parts that turn extracted values into posted bills: an Azure subscription and resource to provision, a schema to define and maintain, retry and throttling logic, a review queue for the pages the model gets wrong, vendor name normalization so that the same supplier does not arrive under four spellings, duplicate detection, and a mapping from extracted fields into whatever your accounting system expects. On a 10,000 invoice a month workload the $220 API bill is very rarely the expensive part of that project.

That is the calculation worth doing before you provision anything. If your team wants structured invoice data rather than a document AI platform to operate, upload an invoice at the top of this page and compare what comes back against the schema you were about to write.