Azure Document Intelligence Pricing: $10 per 1,000 Invoice Pages, Read Rates, and a No-Code Alternative

Last updated August 2026. Rates re-verified 30 August 2026 against Microsoft's Retail Prices API, and every meter, commitment tier and overage rate on this page was unchanged. Azure's own Document Intelligence pricing page still renders every rate as a dash rather than a number and directs you to sign in to the pricing calculator, so the figures on this page now come from Microsoft's public Retail Prices API at prices.azure.com instead, which is the billing source of truth. See the full live rate card further down, including the commitment tiers that never appear on the pricing page. Confirm your own numbers against your agreement before you build.

Azure Document Intelligence pricing vs InvoiceExtractor, side by side

Document Intelligence prices a raw OCR API per page, while InvoiceExtractor prices a finished, no-code product per month. The headline rates measure different things: one is the cost of a single model call, the other is the cost of a working invoice-to-spreadsheet workflow with nothing to build. Here is an honest comparison on the things that actually decide your total cost.

What the prebuilt Invoice model really costs per invoice

On the pricing page the prebuilt Invoice model looks almost free: $10 per 1,000 pages, about a penny a page. Unlike some cloud APIs, Azure does not round each document up to a block, so a one-page invoice really is about $0.01. Run 2,500 invoice pages a month and the base API charge is roughly $25, and 10,000 pages is about $100. The numbers that do not show on the headline are the free-tier limits and the add-ons. The free F0 tier reads only the first 2 pages of any file and stops at 500 pages a month, so any real AP volume runs on the paid S0 tier. And if you switch on Query Fields to pull values the model does not return by default, that adds $10 per 1,000 pages, while high-resolution, formula, and barcode add-ons add $6 per 1,000 each. With a couple of those enabled your effective rate is well above $0.01 a page.

That number is still only the API charge. The Invoice model returns structured JSON with fields such as vendor, invoice number, date, totals, and line items, each with a confidence score. To get that into your accounting system, a developer has to set up an Azure subscription, deploy a Document Intelligence resource, manage keys and endpoints, call the SDK, map every field to your columns, reassemble line items into clean rows, and keep it working as vendor layouts change. If a layout needs a custom model you also pay $30 per 1,000 pages plus $3-an-hour training. Add the engineering hours and the all-in cost for an invoice-only team is usually far above the sticker rate.

Now put it next to a flat plan. InvoiceExtractor Starter is $49 a month and includes 2,500 pages, and the Plus plan is $149 a month for 10,000 pages, both in US dollars, with full line items, multi-page and multi-invoice handling, a review screen, and Excel or CSV output, and no code to write or maintain. At a few thousand invoices a month with the fields you actually need, the no-code plan is competitive with Document Intelligence before you even count the build. If you have engineers and an Azure pipeline already, Document Intelligence can still fit. If you do not, the flat tool is almost always cheaper once you count the add-ons and the engineering. That is the honest trade-off, and it is worth doing the math on your own invoice volume.

Where Azure Document Intelligence is the better choice

Document Intelligence earns its place for the job it was built for. If you have developers working inside Azure, its strong OCR, broad library of prebuilt and custom models beyond invoices, batch processing, and the ability to keep documents inside your own Azure tenant for compliance are real advantages. At very high volume across many document types, the per-page rate is hard to beat, and teams that need fine control over the JSON response and how each field flows into their systems get a well-documented, capable primitive. For raw document OCR beyond invoices, an enterprise platform such as document data extraction software covers broader document types, and Document Intelligence sits comfortably in that build-your-own world.

Where a no-code invoice tool wins

For a US team whose whole job is turning vendor invoices into clean data, wrapping a raw model in custom code adds cost and fragility you do not need, and the add-on fees and free-tier limits work against a simple invoice workflow. InvoiceExtractor reads any invoice layout with no template and no code, captures every line item, splits batches that contain multiple invoices, and exports straight to Excel or CSV. The invoice data extraction software overview covers exactly what gets captured, and the invoice OCR software page shows how the AI reads PDFs, scans, and photos. If you do want a programmatic option, the invoice data extraction API returns the same structured data on a flat plan, without per-page OCR charges. Comparing the wider field on cost? The best invoice data extraction software roundup, the Azure Document Intelligence alternative and AWS Textract pricing breakdowns, and the Google Document AI pricing and Nanonets pricing pages cover the other cloud-API, platform, and developer sides of the market.

Extraction is usually the first step in a larger workflow. Once invoice data is structured, many teams route it straight into approvals and payment, so pricing the whole accounts payable stack is worth doing alongside the extraction tool. A dedicated accounts payable automation platform handles the approval and payment side once the data is clean. Getting the extraction step right first keeps everything downstream fed with reliable data.

You may already be paying for Document Intelligence inside another product

This is the part of the pricing question most buyers miss. Azure Document Intelligence is the OCR engine embedded in other Microsoft products, so a finance team can be consuming it without ever opening the Azure portal. The clearest example is the Payables Agent in Dynamics 365 Business Central: Microsoft documents that the agent sends each PDF invoice attachment to Azure Document Intelligence for extraction and stores the result in the Inbound E-Document record. You never see a per-page line item for it, because Business Central bills the agent through Copilot Credits on consumption-based billing rather than the published Azure page rate.

That matters for a cost comparison in two ways. First, the effective per-invoice cost inside Business Central is not the $10 per 1,000 pages on the Azure pricing page, it is whatever the Copilot Credit consumption works out to, and Microsoft notes charges vary with interaction complexity. Second, the embedded route carries limits the raw API does not: the agent will not process PDFs over 10 pages or over 5 MB, skips emails with more than 10 attachments, and caps at 100 emails and 500 invoices per day. If you are running the numbers for a Business Central shop, our Business Central AP automation page lays out those documented constraints against a flat-rate capture layer, and the guide to importing invoices into Business Central covers the bulk routes the agent does not handle.

The practical takeaway: before you price a Document Intelligence build, check whether the workload is already covered by a product you license, and whether that product's limits fit your invoice volume. Paying twice for the same OCR engine, or building on a raw API when an embedded one would do, are both easy mistakes at this stage.

Is Azure Form Recognizer the same as Azure AI Document Intelligence?

Yes. Microsoft renamed Azure Form Recognizer to Azure AI Document Intelligence, and the service, the models and the pricing are the same thing under a newer name. Older tutorials, SDK package names and Stack Overflow answers still say Form Recognizer, so if you are comparing a quote or an old cost estimate against current rates, you are comparing the same product.

This matters when you search for pricing, because the three names return different pages. "Azure Form Recognizer pricing", "Azure AI Document Intelligence pricing" and "Azure Document Intelligence pricing" all describe the same rate card: $10 per 1,000 pages for the prebuilt Invoice model, $1.50 per 1,000 for Read OCR, $30 per 1,000 for custom extraction, and an F0 free tier capped at 500 pages a month and the first 2 pages of any file. The naming change did not alter what you pay per page, only what the docs call it.

One practical consequence: if your team built an integration against the Form Recognizer SDK, the client libraries were renamed too, and the API version you pin determines which model set you can call. Older API versions do not expose newer add-ons such as Query Fields, so an integration that has not been touched in a couple of years may be missing capabilities you are budgeting for. Confirm the API version before you cost out a rebuild.

Why the Azure pricing page shows a dash instead of a price

If you have opened the Document Intelligence pricing page recently and found no numbers on it, you are not misreading it. As of 20 August 2026 every rate on that page renders as a dash, with a note that prices are estimates only and an instruction to sign in to the Azure pricing calculator to see pricing based on your current program or offer. The only concrete figure still printed is the free tier: 500 pages a month at no cost, web or container.

That is a meaningful change for anyone building a business case. Azure rates are now effectively account-specific, which means a per page number quoted in a blog post, a vendor comparison or an AI assistant answer is a historical figure rather than your price. Pull your own rate from the calculator while signed in, and treat every published number, including the ones on this page, as a benchmark to sanity check it against.

The same invoice model through Power Platform costs about eight times more

Microsoft sells this extraction through two different front doors at two very different prices, and a lot of teams pick the expensive one without realizing there was a choice. AI Builder, the AI layer inside Power Apps and Power Automate, exposes a prebuilt invoice model that returns the Document Intelligence prebuilt invoice schema field for field, down to the Dutch KVKNumber and the Australian BPay biller code. Same fields, same job. It just meters in credits rather than pages.

The premium is not indefensible. Going through Power Platform gets you the flow designer, Dataverse, tenant governance and an approval process you did not have to build, and for a departmental workflow reading a few dozen documents a month that is worth more than the page rate. What catches teams out is scale: the credit meter is linear while the workflow value is not, so at a few thousand pages a month you are paying a large multiple for convenience you already stopped needing.

Two dates matter if you are on that path. Microsoft closed AI Builder capacity add-ons to new customers on 1 November 2025, and on 1 November 2026 it removes the AI Builder credits seeded in Power Apps Premium, Power Automate Premium, Dynamics 365 Finance and Operations and the rest, with no conversion to Copilot Credits. The full credit rate card and what each license actually seeds is broken down on our AI Builder pricing page, and the head to head is covered in AI Builder vs Azure Document Intelligence.

The live Azure Document Intelligence rate card, from Microsoft's Retail Prices API (verified 26 August 2026)

Azure's public pricing page for Document Intelligence still renders every rate as $-. We have now checked it on four separate occasions across 2026 and it has never once served an actual number to a plain HTTP request. The prices are injected client-side, which means crawlers, answer engines and anyone reading the page without JavaScript get nothing.

So we stopped asking the marketing page. Microsoft operates a public, unauthenticated billing API at prices.azure.com/api/retail/prices, and it returns the rates Azure actually bills from. Everything below was pulled from it on 23 August 2026 for the East US region. These are live figures, not a cached estimate.

Pay-as-you-go (S0), per 1,000 pages

Batch meters exist for Read, Layout, Prebuilt, Custom, Add-on, Query fields and Classifier, and every one of them is priced identically to its synchronous equivalent. Batching on Azure buys you throughput, never a discount. That is worth knowing because several competing services do discount batch work: Mistral OCR halves to $2.00 per 1,000 pages in batch, and Anthropic and OpenAI both take 50% off batch jobs.

Read is the only meter with a volume break, and it needs 1,000,000 pages in a month before the rate drops from $1.50 to $0.60. Prebuilt invoice processing has no pay-as-you-go volume discount at all.

Commitment tiers: the discount that is not on the pricing page

If you have predictable invoice volume, Azure sells prebuilt processing on monthly commitment tiers. These are entirely absent from the rendered pricing page, so most buyers never find out they exist.

Overage on every tier is billed at that tier's effective rate, so going slightly over your commitment costs the same per page as staying inside it. That is unusually forgiving, and it means the only real risk of committing is under-using the tier you bought.

Disconnected containers, the option for air-gapped environments, are billed annually: $8,640 per year at 100K, $38,400 at 500K, and $72,000 at 1M. Divide those by twelve and they come to $720, $3,200 and $6,000 a month, which is exactly the connected container price. Microsoft charges no premium whatsoever for running Document Intelligence fully disconnected. For a regulated US buyer with an air-gap requirement, that is a genuinely useful thing to know, and it is nowhere on the pricing page.

How that compares

The gap between $0.16 and $10.00 is not a pricing anomaly, it is a description of scope. The cheap rows hand you text and leave the invoice schema, the line item reconstruction and the validation to you. Azure and AWS hand you a typed invoice object with an SLA behind it. Full rate cards for the cheap end are on our Qwen OCR pricing and AWS Textract pricing pages.

The commitment tiers Microsoft never puts on the pricing page

Pay-as-you-go is the only thing the pricing page shows, but the Retail Prices API carries a full set of commitment tiers for every meter that matters to document work. Re-read on 26 August 2026, the rates below have not changed since our previous check, and a 20,000 page prebuilt commitment still beats pay-as-you-go by 24%. Twenty thousand pages a month is a mid-market AP department, not an enterprise.

Custom extraction has its own tier ladder, and it is the one nobody publishes. Pay-as-you-go custom extraction is $30 per 1,000 pages. A commitment cuts that to between $27 and $18 depending on volume, and running the model in a connected container cuts it again by a further 15%.

Disconnected custom extraction, the air-gapped option, is billed yearly at $24,480 for 100,000 pages a month, $107,100 for 500,000 and $183,600 for a million. Divide those by twelve and they land at $2,040, $8,925 and $15,300 a month, which is exactly the connected container price. Microsoft charges no premium at all for running Document Intelligence in an air-gapped environment, and that holds for the prebuilt models too.

Read, the plain OCR meter, has a separate ladder that only opens at 500,000 pages a month. It is the cheapest per-page commitment in the whole service.

Azure Document Intelligence versus an on premises capture license

Azure prices prebuilt invoice extraction at $10.00 per 1,000 pages with no upfront commitment. The traditional alternative is a perpetual capture license installed on your own servers, and the two models behave very differently at low volume. On OpenText published SKU price list, an Intelligent Capture Standard Server Bundle with Advanced Recognition covering 100,000 pages a year lists at $10,798, plus $4,296 for every attended client seat, plus 23 percent first year maintenance that then rises 10 percent a year.

The cloud meter wins decisively on cost until the requirement is physical mail, scanners, checks and an operator verification workflow, which is what an enterprise capture platform is actually for. The full breakdown is on our OpenText Intelligent Capture pricing page, including the maintenance uplift that most quotes never spell out.