What is Azure Document Intelligence?
Azure Document Intelligence in Foundry Tools, formerly Form Recognizer and then Azure AI Document Intelligence, is Microsoft's cloud service for reading documents. The Read model pulls out text, Layout adds tables, paragraphs, selection marks and structure, prebuilt models return fields from invoices, receipts, ID documents, US tax forms, health insurance cards and contracts, and you can train custom extraction and classification models on your own labelled samples. The current API is v4.0 (2024-11-30), which is generally available with no announced retirement date.
Microsoft now describes it as part of Azure Content Understanding, its newer generative-model service, and says in an October 2026 guide that working Document Intelligence workloads do not need to migrate. The two keep separate APIs, SDKs and billing.
Pay-as-you-go prices in East US are $1.50 per 1,000 pages for Read (falling to $0.60 after the first million pages a month), $10 for Layout and prebuilt models, $30 for custom extraction and $3 for custom classification. A free tier covers 500 pages a month, and Docker containers, including fully disconnected ones on annual commitments, let you run it on your own hardware.
How much does Azure Document Intelligence cost?
| Plan | Price | What you get |
|---|---|---|
| Free (F0) | Free | 500 pages a month, first two pages of each document |
| Read | $1.50 per 1,000 pages | $0.60 per 1,000 above 1 million pages a month |
| Layout and prebuilt models | $10 per 1,000 pages | |
| Custom extraction | $30 per 1,000 pages | |
| Custom classification | $3 per 1,000 pages | |
| Commitment tiers | From $190/month | Prepaid page bundles at a discount |
| Disconnected containers | From $8,640/year | Annual commitment, runs fully offline |
Free tier 500 pages a month; Read $1.50 per 1,000 pages ($0.60 above 1 million a month); Layout and prebuilt $10; custom extraction $30; classifier $3; add-ons $6; commitment tiers from $190/month; disconnected containers from $8,640/year.
Is Azure Document Intelligence worth it?
Somebody in your finance team still types invoice totals into a spreadsheet, and every few weeks a typo costs an afternoon to find.
Azure Document Intelligence reads the invoice and hands back the vendor, date, line items and total as data. It comes with ready-made models for invoices, receipts, passports and driving licences, US tax forms and health insurance cards, so for common paperwork you start without training anything. For your own forms, you label a handful of samples and train a custom model. Training time is free up to 10 hours a month and $3 an hour after that.
Plain text is where the price is hard to beat. Read costs $1.50 per 1,000 pages and $0.60 after the first million a month, against $4 for Mistral OCR and $10 for Reducto's parser. Layout and the prebuilt models cost $10 per 1,000 pages, the same as Reducto's Parse.
The catch: the product is built for documents that look alike, and Microsoft now says as much. Its October 2026 guide tells you to keep a working Document Intelligence pipeline, but to look first at Content Understanding, its newer service, for new cloud OCR projects, for varied layouts, for fields that have to be inferred, and for preparing documents for an AI chatbot. Costs also stack: a single page can be billed for a prebuilt model plus $6 for add-ons such as high-resolution reading and formulas, and the service reads every page unless you tell it which ones. Office files work with Read and Layout only, not the prebuilt or custom models.
Our take: think of it as the reliable office scanner, not the clever assistant. If your documents are the same handful of forms every month, or must never leave your building, it is hard to beat on price and predictability. For a pile of different contracts, a generative parser will do more of the reading for you.
Where it earns its place
An accounts-payable team of six at a 500-person distributor
Stop typing supplier invoices into the ERP by hand
The prebuilt invoice model returns vendor, totals and line items as JSON for $10 per 1,000 pages; 20,000 invoices a month costs about $200 pay-as-you-go, or $190 on the 20,000-page commitment tier.
A platform engineer at a defence contractor
Run OCR on classified scans with no internet connection
Disconnected containers run the service fully offline on your own hardware under an annual commitment, starting at $8,640 a year for 100,000 prebuilt pages a month. Microsoft's guide names air-gapped work as a case for Document Intelligence over Content Understanding.
A data engineer digitising a 10-million-page records archive
Make old scanned files searchable as cheaply as possible
Read alone charges $1.50 per 1,000 pages for the first million each month and $0.60 after that, so most of the archive bills at the lower rate. Skip Layout unless you need tables, since it costs $10 per 1,000.
Before you commit
- Setup
- Create a Document Intelligence resource in the Azure portal, take the key and endpoint, and call the REST API or a client SDK; Document Intelligence Studio lets you try models and label custom training data in a browser.
- Integrations
- REST API and client SDKs, Document Intelligence Studio, Azure Blob Storage for training data and batch jobs, and containers for on-premises and disconnected use. Input is PDF, JPEG, PNG, BMP, TIFF and HEIF, plus DOCX, PPTX and XLS for Read and Layout.
- Data and privacy
- Documents are processed in the region of your resource; submitted input and results are kept for 24 hours and then deleted automatically, or immediately through the Delete Analyze Result API.
- Cost at scale
- One million prebuilt pages a month costs $10,000 pay-as-you-go or $7,500 on a commitment tier; custom extraction on the same volume is $30,000, or $18,000 committed.
- Limits
- Standard tier: 500 MB and 2,000 pages per document, 15 analyse requests per second by default (raisable through support). The free tier reads only the first two pages of each document, up to 4 MB.
What's good
- Read OCR at $1.50 per 1,000 pages, $0.60 above a million
- Ready-made models for invoices, receipts, IDs and US tax forms
- Custom models from a few labelled samples, with 10 free training hours a month
- Docker containers, including fully offline ones
- Documents and results deleted 24 hours after processing
What's not
- Microsoft steers new cloud OCR projects to Content Understanding
- Struggles with varied layouts and fields that need inference
- Add-ons and multiple models stack on the per-page price
- Office files supported only by Read and Layout
- Default limit of 15 analyse requests per second
Best for
Teams processing high volumes of standard forms, invoices or IDs on Azure, or anyone who needs document extraction to run on-premises or offline.
Azure Document Intelligence alternatives
Azure Document Intelligence: common questions
- Is Azure Document Intelligence worth it?
- For invoices, receipts and forms that look the same every time, Azure Document Intelligence is cheap, steady and runs offline, but it is the wrong tool for messy documents that need judgement. We rate it 4.0 out of 5.
- How much does Azure Document Intelligence cost?
- Free tier 500 pages a month; Read $1.50 per 1,000 pages ($0.60 above 1 million a month); Layout and prebuilt $10; custom extraction $30; classifier $3; add-ons $6; commitment tiers from $190/month; disconnected containers from $8,640/year.
- What are the downsides of Azure Document Intelligence?
- Microsoft steers new cloud OCR projects to Content Understanding; struggles with varied layouts and fields that need inference; add-ons and multiple models stack on the per-page price; office files supported only by Read and Layout; default limit of 15 analyse requests per second.
- Who is Azure Document Intelligence best for?
- Teams processing high volumes of standard forms, invoices or IDs on Azure, or anyone who needs document extraction to run on-premises or offline.
- How much does Azure Document Intelligence cost for a team or in production?
- One million prebuilt pages a month costs $10,000 pay-as-you-go or $7,500 on a commitment tier; custom extraction on the same volume is $30,000, or $18,000 committed.
- Is Azure Document Intelligence safe for company data?
- Documents are processed in the region of your resource; submitted input and results are kept for 24 hours and then deleted automatically, or immediately through the Delete Analyze Result API.
- What are the best Azure Document Intelligence alternatives?
- The closest alternatives we have reviewed are Reducto, LlamaParse and Mistral OCR.
Answers checked Oct 26, 2026.
Make Azure Document Intelligence? Show this rating on your site
