UiPath Document Understanding vs Azure Document Intelligence Pricing

Short answer: Azure AI Document Intelligence charges $10 per 1,000 pages for its prebuilt invoice model. UiPath Document Understanding charges $8,000 for a 40,000 page bundle, which is $200 per 1,000 pages, and that bundle only runs if you also license Studio, Orchestrator and at least one robot. On the extraction step alone, UiPath costs 20 times more. It is also not selling the same thing, and for some buyers the difference is worth every dollar. Here is the arithmetic on both sides.

These two land on the same shortlist for a specific reason. A finance or shared services team decides it wants invoices read automatically, IT gets involved, and IT already has a Microsoft agreement or an RPA program. From there the evaluation becomes a cloud API against an automation platform, and nobody sets the comparison up in a way that makes the price difference legible. Both vendors are partly to blame: Azure's pricing page renders empty rates in a browser, and UiPath does not publish a Document Understanding price at all.

Where each price actually comes from

Azure publishes its rates through the Retail Prices API even when the pricing page shows nothing useful. Pull the Document Intelligence meters for East US and the S0 pay as you go card is complete and unambiguous.

UiPath does not have an equivalent public rate card, but it does have a public price list. The UiPath GSA Pricebook is a US federal schedule document, and it carries the Document Understanding family with list price and discounted federal price printed side by side.

The cloud editions of all three carry identical prices to the on premises editions, to the cent. Even at the ten million page tier, which requires committing a million dollars a year, UiPath is still ten times the Azure prebuilt invoice rate. At the volume an actual accounts payable department produces, it is twenty times.

The comparison people get wrong: the bundle is not the bill

An Azure quote for invoice extraction is one line. You call the prebuilt invoice model, you are billed $0.01 per page, and there is nothing else to buy. A UiPath quote never looks like that, because Document Understanding is a service inside a platform rather than an endpoint you call.

On the same federal price list, the minimum realistic configuration for running Document Understanding on premises is Studio at $3,000, Orchestrator at $20,000, one unattended robot at $8,000, and the 40,000 page bundle at $8,000. That is $39,000 a year at list, and Premium Care support carries a $12,000 floor on top of its 12 percent rate. The cloud path is lighter, at roughly $22,840 for a developer seat, a cloud unattended robot, an Action Center seat and the same page bundle.

That is the honest shape of it. Reading 40,000 invoice pages costs $400 on Azure and somewhere between $8,000 and $51,000 on UiPath depending on what you already own. If you are an existing UiPath shop with Studio, Orchestrator and robots already paid for, the incremental cost is the $8,000 bundle and the comparison narrows to 20x. If you are starting from nothing, it does not.

Both vendors meter, and both meters have traps

Neither product bills a flat per page rate once you look closely, and the traps are different.

UiPath sells pages but consumes units. Classification costs 0.2 AI Units, extraction after classification costs 0.8, and extraction without a classification step costs a full 1.0. A page that is classified and extracted is capped at 1 unit, which is how the arithmetic is meant to work. But run two extractors over the same page and the second is charged again at 0.8. Turn on generative validation, which automatically checks low confidence fields instead of routing them to a person, and the page costs 2 units. A 40,000 page bundle then covers 20,000 real pages and the effective rate doubles to $400 per 1,000. UiPath's own SKU description warns about this in one line: certain features and models may be metered at different rates.

Azure's trap is the opposite. Its rate card is flat and predictable, but the meter you pick changes the price by a factor of twenty and the names do not signal it. Read is $1.50 and returns text with no idea which number is the total. Layout is $10 and returns structure without invoice semantics. The prebuilt invoice model is also $10 and returns typed fields plus line items, which is the one you almost always want. Custom extraction is $30 and only pays for itself if your documents genuinely do not fit the prebuilt schema. Teams routinely train a custom model at triple the price to solve a problem the prebuilt model already solved.

One more Azure detail worth knowing, because Microsoft publishes it nowhere obvious: prebuilt commitment tiers exist. Committing to 20,000 pages a month costs $190 a month, which is $9.50 per 1,000. At 100,000 pages a month it is $900, or $9.00. Run it in a connected container instead and the same 20,000 page tier drops to $152 a month, $7.60 per 1,000, a 24 percent saving over pay as you go at a volume a mid sized AP department already hits. There is also no batch discount on any Azure Document Intelligence meter, which is unusual, since Mistral, OpenAI and Anthropic all halve their batch rates.

What UiPath gives you that Azure does not

Price is the wrong place to end this comparison, because the two products stop at different points in the process.

Azure returns JSON. That is the whole product. Something else has to receive that JSON, decide whether the confidence is high enough, show a person the ones that are not, match the invoice against a purchase order, and post it. All of that is your build.

UiPath ships most of that. Validation Station and Action Center give a non technical reviewer a field by field correction screen with the document image beside it, wired into an approval chain, and neither meters. A single Document Understanding project can classify a mixed inbound stream and route invoices, purchase orders and remittance advices to different extractors. And because the extraction sits inside an orchestrated robot, the same process can log into a supplier portal, download the PDF, and post the result into an ERP that has no API at all. That last capability is not something a per page endpoint can substitute for at any price.

UiPath also charges nothing for several things buyers assume cost money. Every OCR engine meters at zero, including UiPath Document OCR, Omnipage, Tesseract, Microsoft OCR and Google Cloud Vision. Model training and serving are free, so retraining on your own vendor layouts costs nothing. On classic projects, RegEx extractors are free and Form Extractors are 0.04 units against 0.2 for the machine learning extractor, which makes routing your highest volume vendors through templates a genuine optimization.

Accuracy, since price without accuracy is meaningless

The most useful published third party test on scanned invoices, run by BusinesswareTech in January 2025, measured Azure AI Document Intelligence at 93.0 percent on header fields and 87.0 percent on line items. AWS Textract came in at 78.0 percent and 82.0 percent. UiPath was not in that test, and no comparable independent benchmark of Document Understanding on US invoices exists in the open, so anyone quoting a UiPath accuracy figure at you is quoting a vendor slide.

What the numbers that do exist show is that line items are where these systems separate. Header fields are close to solved. Table rows, with merged cells, continuation pages and inconsistent column orders, are not. If your evaluation only checks that the vendor name and total came out right, it will not tell you which product to buy. Run your ten worst vendor layouts through both and count the line items.

Whatever you choose, exceptions do not disappear, they move. Someone still opens the queue and fixes the fields the model was unsure about, and the teams that get value from either platform treat that as a back-office operations workflow with owners, queues and a target turnaround, rather than as a folder somebody checks when they remember. A 93 percent field accuracy rate still leaves seven pages in a hundred needing a human, and how fast that human works decides your real cost per invoice more than the license does.

How to choose

Choose Azure AI Document Intelligence if the requirement is invoice data as JSON, you have engineers who will handle the surrounding workflow, and you want a rate you can predict to the dollar. At $10 per 1,000 pages with commitment tiers available at volume, it is one of the cheapest credible invoice extraction services on the market, and its published line item accuracy is the best of the major clouds.

Choose UiPath Document Understanding if you already run UiPath, or if the document step is one node in a longer automation that has to drive user interfaces, or if the human review workflow matters more to you than the per page rate. In an existing estate the incremental cost is a single $8,000 SKU, and at that margin it is a reasonable price for a supported enterprise service with a review UI attached.

Choose neither if the honest requirement is "stop retyping invoices into the ledger". That does not need a platform or a cloud subscription, and it does not need engineers. Our own plans run $49 and $149 a month, or $288 and $888 billed annually, with line items extracted rather than header fields only. Price all three before the proof of concept and the gap between them will tell you which problem you actually have.

The full breakdown on each side is worth reading before a quote lands: UiPath Document Understanding pricing carries every SKU, both metering tables and the federal discount rates, and Azure Document Intelligence pricing carries the full meter list plus the commitment and disconnected container tiers Microsoft does not show. For the broader view of what invoice OCR costs across every vendor, OCR for accounts payable normalizes them all to a cost per 1,000 pages.

Last updated August 2026. Azure rates read from the Microsoft Retail Prices API for East US. UiPath SKU prices transcribed from the public UiPath GSA Pricebook; metering rates from current UiPath Document Understanding documentation. Accuracy figures from the BusinesswareTech scanned invoice test, January 2025. Vendor pricing changes; verify against the source before you commit to a contract.