Last updated September 2026. Intelligent document processing software is sold two completely different ways, and comparing a price from one model against a price from the other is how procurement decisions go wrong. The cloud APIs publish a per-page rate you can check in a browser. The platform vendors publish nothing and quote you a seat price. This page puts the published numbers in one table, taken from each vendor's own price list rather than from a review site, and works out what a real monthly volume actually costs.
The single most useful thing to know before reading further: plain OCR costs exactly the same on all three major clouds, $1.50 per 1,000 pages. Amazon, Microsoft and Google have converged on the identical number, and so has basic table structure, where Azure and Google both charge $10.00. The prices diverge violently at one specific capability, form and key-value extraction, which runs $10.00 on Azure, $30.00 on Google and $50.00 on AWS. That is a five times spread on the step most invoice and statement projects depend on.
The two pricing models, and why they do not compare
Intelligent document processing, usually shortened to IDP, covers any system that reads a document, works out what it is, pulls structured fields out of it and hands them to another system. The category includes cloud APIs from the hyperscalers, specialist SaaS platforms, and document understanding modules bolted onto enterprise automation suites.
Cloud APIs bill per page or per document, with no floor and no contract. Send 400 pages, pay for 400 pages. Nothing is committed, and a page next month costs what it cost this month. That model suits variable volume and a team that has an engineer.
Platform vendors bill per seat, per workflow or per volume band, almost always after a sales call. Some publish an entry price and go quiet above it. Their pitch is that the price includes the parts the raw API does not: a review interface for the humans who correct low-confidence fields, a training loop, connectors into your ERP, audit trails and named support. Whether that is worth the premium depends on whether you would otherwise pay a developer to build those things.
The mistake to avoid is anchoring on a per-page number and then being shocked by a platform quote. A $0.01 page rate is the cost of the model call. A platform price is the cost of the model call plus everything around it.
Published per-page rates from the three major clouds
These are US region list prices for pay-as-you-go usage, read from each vendor's own published price list. All figures are per 1,000 pages.
| Capability | AWS Textract | Azure Document Intelligence | Google Document AI |
|---|---|---|---|
| Plain OCR, text detection | $1.50 | $1.50 | $1.50 |
| Layout and table structure | $15.00 (Tables) | $10.00 (Layout) | $10.00 (Layout Parser) |
| Form and key-value extraction | $50.00 (Forms) | $10.00 (prebuilt models) | $30.00 (Form Parser) |
| Custom trained extraction | $25.00 (Custom Queries) | $30.00 | $30.00 |
| Document classification | Not a separate meter | $3.00 | Included in the parser |
| Natural-language queries on a page | $15.00 | $10.00 (query fields) | Not sold as a meter |
| Rate above the first volume tier | $0.60 OCR above 1M pages | Commitment tiers, up to 60% off | $0.60 OCR above 5M, $20.00 Form Parser above 1M |
Two rows deserve a warning. AWS charges $50.00 per 1,000 pages for its Forms feature, the most expensive meter in this table by a wide margin, and it is easy to switch on by accident because the same API call takes Tables and Forms as feature flags on one request. And Azure's Layout meter at $10.00 is the cheapest published route to structured tables anywhere in this comparison, tied with Google's Layout Parser and a third cheaper than AWS.
What 10,000 pages a month actually costs
Rates per thousand are hard to feel. Here is the same workload, 10,000 pages a month, priced at list on each platform for the two jobs buyers actually run.
| Job | Azure Document Intelligence | AWS Textract | Google Document AI |
|---|---|---|---|
| Plain OCR only | $15.00 | $15.00 | $15.00 |
| Table and layout structure | $100.00 | $150.00 | $100.00 |
| Form and key-value fields | $100.00 | $500.00 | $300.00 |
| Annual cost of the form job | $1,200.00 | $6,000.00 | $3,600.00 |
Three things fall out of that. Choosing the wrong cloud for form extraction costs $4,800 a year at a genuinely modest volume, which is more than most small finance teams spend on software in total. Plain OCR is ten to fifty times cheaper than extraction, so the highest-value optimization available is usually deciding which pages actually need structure and routing the rest through the cheap meter. And on the table job the three clouds are within $50 a month of each other, so if tables are all you need, price is not the deciding factor and you should choose on accuracy with your own documents.
The costs that do not appear in the rate card
Per-page pricing looks clean until you deploy it. Three categories of spend hide outside the meter.
Idle infrastructure. Google charges $0.05 per hour to keep a deployed processor running, which is $36.50 a month whether or not you send it a single page, and provisioned throughput is $300.00 a month on the same basis. Azure's connected containers and disconnected deployments carry their own commitments. Run several processors for several document types and the floor alone can exceed your usage bill.
Commitment tiers that are barely discounts. Azure's prebuilt commitment tier starts at $190.00 a month for 20,000 pages. Those pages cost $200.00 at the pay-as-you-go rate, so the commitment saves 5 percent and locks you in for the privilege. The real savings sit much higher up the curve: Azure's Read commitment at 2 million pages is 60 percent off. Do the arithmetic against your own volume instead of assuming a commitment tier is cheaper.
Human review. Every honest IDP deployment budgets for a person checking low-confidence extractions. On invoices and statements, 95 percent field-level accuracy sounds excellent and still means one field in twenty needs a human. At 10,000 documents a month with fifteen fields each, that is 7,500 corrections. Whatever those cost in salary usually dwarfs the API bill, and it is the number the platform vendors are really selling against.
What the specialist platforms charge
Here the honest answer is that most of them do not say. Vendors including Rossum, Hyperscience, Instabase and the document understanding modules inside the large RPA suites price by quote, and any specific figure in a comparison article was either one customer's contract or an invention. We are not going to reprint numbers we cannot verify.
What is publicly checkable is the shape. Specialist platforms typically bill on document volume bands rather than pages, add for connectors, extra document types and named support, and cap the number of custom document types on entry tiers. Enterprise agreements are annual and often carry a minimum commitment that survives a drop in your volume, which is the term to negotiate hardest if your document flow is seasonal.
Google is the exception worth studying, because it publishes per-document prices for specialized parsers and they sit on a completely different axis from its per-page rates: invoice $0.10 per document, expense and receipt $0.10, pay slip $0.30, W2 $0.30, US driver license $0.10, and bank statement $0.75. Normalising those against Form Parser at $30.00 per 1,000 pages, which is $0.03 a page, gives you a break-even. A bank statement is worth the $0.75 specialized parser only above roughly 25 pages; below that, page pricing is cheaper. An invoice breaks even near three pages, so on typical one and two page invoices the general parser costs less per unit, and you are paying the $0.10 for the invoice-specific field model rather than for the reading.
How much does intelligent document processing software cost?
For cloud APIs, between $1.50 and $50.00 per 1,000 pages depending on what you extract: plain OCR is $1.50 across all three major clouds, table structure is $10.00 to $15.00, and form and key-value extraction is $10.00 to $50.00. For specialist platforms, expect quote-based annual contracts that start in the low hundreds of dollars per month and scale with document volume, connectors and support level. Budget human review separately, because it usually costs more than the software.
What is the difference between OCR and intelligent document processing?
OCR turns pixels into characters. Intelligent document processing works out what the document is, which of those characters are the invoice total, and where that total should be written. OCR is one step inside IDP and it is the cheapest step: $1.50 per 1,000 pages against $10.00 to $50.00 for the extraction layer on top. A tool that only does OCR hands you text, not fields, and the gap between those two things is most of the project.
Is AWS Textract cheaper than Azure Document Intelligence?
For plain OCR they are identical at $1.50 per 1,000 pages. For tables Azure is cheaper, $10.00 for Layout against $15.00 for Textract Tables. For forms and key-value pairs the gap is dramatic: $10.00 on Azure against $50.00 on Textract, a five times difference. On published list pricing Azure is the cheaper of the two for every structured extraction job, and the gap only narrows above a million pages a month where Textract's automatic tiering kicks in.
Do I need IDP software or a PDF converter?
Count your document layouts and name your destination. A handful of consistent layouts landing in a spreadsheet is a converter job: tens of dollars a month, running this afternoon. Many varying layouts landing in an ERP with approvals and an audit trail is an IDP job and takes a project. The cost difference between those two answers is roughly a hundred times, so it is worth ten minutes to get the diagnosis right.
Can intelligent document processing handle scanned documents?
Yes, and it is the main reason the category exists. Every platform here runs OCR first on pages with no text layer, then applies extraction to the recognised text. Accuracy tracks scan quality closely, so 300 dpi flatbed scans behave very differently from a photograph of a page taken on a phone at an angle. Price the human review budget against your worst input quality, not a clean sample.
What is the cheapest way to extract tables from PDFs at volume?
On published list pricing, Azure Document Intelligence Layout and Google Layout Parser tie at $10.00 per 1,000 pages, a third less than AWS Textract Tables at $15.00. But the raw rate is rarely the whole cost. Add engineering time to build against the API, idle hosting where it applies, and the review step. Below a few thousand pages a month, a fixed-price converter is almost always cheaper than any per-page API once the build is counted.
When intelligent document processing is the wrong purchase
IDP earns its cost when documents arrive in many formats, need classifying before extraction, feed a system of record, and carry an audit requirement. Accounts payable at scale is the textbook case, and so is claims intake in insurance.
It is the wrong purchase when the job is narrower. If what you have is a stack of PDFs with tables in them and a spreadsheet at the other end, you do not need a classification model, a training loop or a human-in-the-loop review queue. You need a converter. The whole IDP apparatus exists to absorb variability, and a monthly statement in the same layout every month has no variability to absorb. One or two layouts into a spreadsheet is a PDF to Excel conversion problem. Twenty layouts into NetSuite with an approval chain is an IDP problem. Buying the second when you needed the first is the most expensive mistake in this market, and it is common because the sales motion is far stronger at the enterprise end.
Once the data is out, the remaining work is plumbing rather than intelligence, and it is a separate budget line: getting extracted fields into the systems that consume them means connecting your applications, APIs and databases reliably enough that nobody has to re-key anything downstream.
Where to go next
For the full first-party rate card on a single vendor rather than the comparison, we have detailed breakdowns of Amazon Textract pricing, Azure Document Intelligence pricing and Google Cloud Vision and Document AI pricing, each built from the vendor's own published price list. On the platform side, our comparisons of Nanonets, Docsumo and Amazon Textract as alternatives cover what each is genuinely good at. If the count-your-layouts test pointed you at the simpler answer, PDF to Excel for enterprise teams is the right starting page.