Azure AI Content Understanding vs AWS Textract: Invoice Extraction Pricing Compared

Short answer: for turning a vendor invoice into structured fields, AWS Textract Analyze Expense costs $10.00 per 1,000 pages and Azure AI Content Understanding field extraction costs $22.00 per 1,000 pages. Textract is 55% cheaper for that job and runs in every major US region, while Content Understanding field extraction is sold in only one US region. Content Understanding wins decisively on plain reading, where basic OCR is $1.00 per 1,000 pages against Textract's $1.50, and on digital files like Word, Excel and email, which it bills at $0.01 per 1,000 pages.

Both vendors make this comparison harder than it should be. AWS stopped rendering prices on its Textract pricing page, which now returns template placeholders instead of numbers. Azure does the same on its Content Understanding page. Every figure below was read from each vendor's own pricing API, the AWS Price List API and the Azure Retail Prices API, so these are the rates your invoice will actually carry. If you would rather skip the comparison and see what structured invoice data looks like, drop a PDF into the extractor at the top of this page.

Azure Content Understanding vs AWS Textract pricing, side by side

The two services split into the same two jobs: reading a document, and returning typed fields from it. Pricing them against each other only works job by job, because the cheaper service flips depending on which one you need.

The headline for an accounts payable team is the fifth row. Analyze Expense is a purpose-built invoice and receipt model that returns vendor, dates, totals and line items for $10.00 per 1,000 pages. Content Understanding gets to the same place through a generative model running against a schema you define, and charges $22.00 for it. On 10,000 invoices a month that is $100 against $220. On 100,000 it is $1,000 against $2,200.

Why Content Understanding costs more for invoices but less for everything else

The gap is architectural rather than commercial. Analyze Expense is a fixed model that AWS trained once on expense documents and serves at scale, so the marginal cost of a page is low and the price reflects it. The tradeoff is that you get the fields AWS decided to extract, and nothing else.

Content Understanding runs a generative model against a schema you write. That is why it can read a document type nobody trained it on, pull a field that exists only on your supplier's layout, and handle audio and video in the same pipeline. It is also why a page costs more than twice as much. You are paying for flexibility you may not need if a standard vendor bill is all you ever process.

The same logic explains the rows where Azure wins. Content extraction at $1.00 per 1,000 pages undercuts Textract's $1.50, and the standard tier bundles tables into the $5.00 rate where AWS charges $15.00 for its Tables feature alone. If a meaningful share of your documents arrive as Word files, Excel attachments or emails with a text layer already present, the $0.01 meter is in a different league entirely: AWS has no equivalent and bills those as ordinary pages.

The regional constraint that decides this for many US teams

Before price enters the conversation, check whether you can buy the Azure service at all. Content Understanding content extraction runs 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, the meter that returns invoice fields, runs in exactly three regions worldwide: West US, Sweden Central and Australia East.

West US is the only one in the United States. If your tenant and your data residency policy put you in East US, which covers a large share of US Azure accounts, you can OCR an invoice on Content Understanding but you cannot ask it to return the invoice number. Getting typed fields means provisioning in West US and moving documents across regions, which is a security review before it is a budget line.

Textract has no such restriction. Analyze Expense is available across the standard US regions, and sync and async calls are priced identically for every feature, so there is no architectural penalty for choosing one calling pattern over the other. Content Understanding charges exactly 50% more for synchronous document extraction, $1.50 against $1.00 on basic and $7.50 against $5.00 on standard, for output that is byte for byte the same.

What about accuracy

Price only matters if the output is usable. The most rigorous published third-party test we have found on scanned invoices, run by BusinesswareTech in January 2025, measured AWS Textract at 78% on header fields and 82% on line items. Azure Document Intelligence, the older Microsoft service, scored 93% on headers and 87% on line items in the same test.

There is no equivalent independent benchmark for Content Understanding field extraction yet, and we are not going to invent one. What the existing numbers do tell you is that Textract's weakness is header fields rather than line items, which is the opposite of the usual assumption, and that an 18% to 22% error rate on either service is enough that a review step is not optional. Budget for someone checking exceptions regardless of which vendor you pick.

Neither price includes the project

Both figures buy an API, not a working accounts payable process. Whichever service you choose, the same work sits on your side of the line: provisioning and securing the cloud resource, defining and maintaining a schema, retry and throttling logic, a review queue for low-confidence pages, vendor name normalization so the same supplier does not arrive under four spellings, duplicate detection, and a mapping from extracted values into whatever your accounting system expects.

On 10,000 invoices a month, the difference between $100 and $220 of API spend is not what decides this project. The engineering time around it is. It is also the part that keeps growing quietly, which is why teams running document pipelines across both clouds usually end up wanting a read-only view of where cloud spend is actually going before the finance conversation gets uncomfortable.

Which one to buy

Buy AWS Textract Analyze Expense if invoices and receipts are the job, you want a flat per-page rate with a volume break above a million pages, and you need US regional coverage beyond West US. At $10.00 per 1,000 pages with line items included, it is the cheaper purpose-built option and the full rate card is on our AWS Textract pricing breakdown.

Buy Azure Content Understanding if you are paying for OCR or table extraction at volume, a real share of your files are digital documents that hit the $0.01 meter, your documents vary so much that no prebuilt schema fits, or you need audio and video in one pipeline. Every meter and its regional availability is on our Azure AI Content Understanding pricing page.

And if you are a Microsoft shop processing ordinary vendor bills, there is a third answer that neither vendor advertises: the older Azure Document Intelligence prebuilt invoice model is $10.00 per 1,000 pages, matches Textract on price, beats both on published accuracy, and offers commitment tiers from $190 a month that Content Understanding does not have at any volume.

Frequently asked questions

Is AWS Textract cheaper than Azure Content Understanding?

For invoice extraction, yes. Textract Analyze Expense is $10.00 per 1,000 pages against $22.00 for Content Understanding field extraction, so Textract is 55% cheaper. For plain OCR the answer reverses: Content Understanding basic extraction is $1.00 per 1,000 pages against $1.50 for Textract Detect Document Text.

How much does Azure Content Understanding cost?

Content extraction is $0.01 per 1,000 pages for digital files, $1.00 for basic OCR and $5.00 for standard layout extraction. Field extraction is $22.00 per 1,000 pages in West US and Australia East, and $20.00 in Sweden Central. There are no commitment tiers and no volume discounts on any meter.

Which Azure regions support Content Understanding field extraction?

Three: West US, Sweden Central and Australia East. West US is the only United States region. Content extraction is much more widely available, running in 16 regions including East US, so a tenant outside West US can read documents but cannot run field extraction on the service.

Does AWS Textract extract invoice line items?

Yes. Analyze Expense returns line item groups alongside header fields such as vendor, invoice number, dates and totals, at $10.00 per 1,000 pages. A common and expensive mistake is using the Forms and Tables features instead, which cost $70.00 per 1,000 pages together and return less invoice structure than Analyze Expense does.

Can I use Content Understanding and Textract together?

Technically yes, and a few teams do split the pipeline, reading documents on the cheaper Azure content meters and extracting fields on Textract. In practice the cross-cloud egress, the second set of credentials and the duplicated review tooling usually cost more than the per-page saving, unless you are running well over 100,000 pages a month.