Azure AI Document Intelligence

What it does

Document Intelligence takes in PDFs, images, and printed or handwritten forms such as receipts, invoices, ID documents and tax forms. It returns structured data: text, key-value pairs, tables, selection marks and document layout. It also offers custom classification and splitting, which sort files into document types and detect boundaries inside multi-document files.

Developers call it through a REST API to feed search indexes, back-end systems and automated workflows such as invoice, claim and receipt processing. Prebuilt models cover common document types, while custom models learn your own layouts from about five sample documents and can be improved with human feedback. Containers allow the same service to run on-premises or at the edge, including on Azure Kubernetes Service. Microsoft now presents it as part of Azure Content Understanding in Foundry Tools, which adds LLM-powered analyzers for more complex or multimodal content.

It suits developers and teams building document-processing pipelines on Azure. Use requires an Azure account. Billing is per page analyzed, with a free tier (F0) that has monthly page limits and does not support premium features. It is a service to integrate, not a ready-made end-user app. Prebuilt models handle only specific document types, and custom extraction needs sample documents and tuning.

What it can do

• Layout extraction: text, tables, titles, paragraphs and selection marks using OCR
• Prebuilt models for receipts, invoices, IDs, W-2 and 1098 tax forms, health insurance cards and contracts
• Custom field extraction trained on your own sample documents
• Custom classification and splitting of multi-document files
• Container deployment on-premises, at the edge or in the cloud

Works with

• Azure Kubernetes Service
• Azure Container Instances
• Azure Stack
• Azure Applied AI Search

Example workflows

Invoice processing

Input: scanned or PDF invoices. Steps: submit them to the prebuilt invoice model through the REST API, then read the returned fields and line items. Output: structured data to pipe into a back-end accounting system.

Custom form extraction

Input: about five sample documents of your own form type. Steps: train a custom extraction model, review its output and add human corrections. Output: a model that returns fields and tables tailored to your layout.

Before you choose

• Billing is per page analyzed, and all pages are processed unless a page range is specified.
• The free tier is limited to 500 pages per month and lacks premium features.
• Custom neural model training is free for the first 10 hours, then $3/hour extra.

Listed under

• Operations
• Data extraction
• Data-extraction
• Developers
• Document processing
• Document-processing
• Enterprise

Get started

Pay as you go, or try Azure free for up to 30 days.

1. Create or sign in to an Azure account.
2. Create a Document Intelligence resource, using the free F0 tier to test.
3. Call the REST API with a prebuilt or layout model, or deploy the container.

Pay-as-you-go per 1,000 pages by model type; no monthly or yearly subscription, though commitment tiers exist for high volume. Container pricing matches cloud pricing. See the Azure pricing page for the rate table.

Source: azure.microsoft.com/en-us/pricing/details/document-intelligence, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Azure Document Intelligence (now part of Azure Content Understanding in Foundry Tools) | Microsoft Azure ↗azure.microsoft.com
• Pricing - Azure Document Intelligence in Foundry Tools | Microsoft Azure ↗azure.microsoft.com

Company website

azure.microsoft.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.