Intelligent document processing pricing: what vendors publish, and what their prices leave out

Intelligent document processing (IDP) pricing comes in four forms. Vendors charge per page, per document, by plan or by quote. Amazon Textract, Azure Document Intelligence and Google Document AI each charge $1.50 per 1,000 pages to read text. Pulling out fields you define costs $15 to $30 per 1,000 pages. Some IDP platforms publish their plans, and others quote each deal after a call. No price list includes the time your team spends fixing results.

We read each vendor's own pricing, product or marketplace pages on October 13, 2026, and we quote only what those pages say. Docsumo, our product, is an IDP platform with a free trial and quoted paid plans.

Four ways IDP vendors charge#

Document AI and IDP price lists use one of four units. The unit decides which of your documents cost the most.

• Per page

  You pay for each page, priced per 1,000 pages. Reading plain text costs the least, and pulling out fields costs more. The three cloud APIs work this way.

• Per document

  Some ready-made models charge per document instead of per page. Google's invoice parser charges $0.10 for each document of up to 10 pages.

• Plans and credits

  A monthly or yearly plan includes a set number of pages or credits, and use beyond it costs extra. Rossum's Starter plan starts at $18,000 a year.

• Quote only

  The vendor's site shows no price, and sales sends a quote after a call. ABBYY Vantage and Tungsten TotalAgility work this way, and so do Docsumo's paid plans.

Open-source OCR, free software that reads the text in a scan, is a fifth option. Tesseract and PaddleOCR are released under the Apache 2.0 license, and the Docling parsing library under the MIT license. All three cost nothing to license. You pay instead for the servers that run them and for the engineers who turn their text into checked fields.

What Textract, Azure and Google charge per 1,000 pages#

The three cloud APIs publish a price for each job. The table shows the price for the first tier of volume each month.

AWS prices are for its US West (Oregon) region, and Microsoft's are for East US. Google's are list prices. Textract's free tier covers 1,000 pages a month of plain text and 100 pages a month of most other features, with none for Custom Queries. Azure also sells query fields, an add-on at $10 per 1,000 pages that adds named fields to a ready-made model.

Prices also change with volume. At 100,000 pages a month of fields you define, Textract's list prices come to $1,500 to $2,500. Custom extraction on Azure or Google comes to $3,000. Azure's 100,000-page commitment tier for custom extraction costs $2,400 a month, paid up front, with extra pages at $24 per 1,000. Google also charges $36.50 a month to host each version of a custom model.

AWS and Google also give custom quotes for large volumes. AWS asks high-volume users to request a custom pricing proposal, and Google's product page offers a custom quote from its sales team. Our comparison of the three services covers what each one sorts and reads, and what it leaves for you to build.

What IDP platforms publish#

Platform price lists vary more, because each platform includes different steps besides reading the text. Some publish a starting plan or a credit price. Others show no price and quote each deal.

Three rows need a note. UiPath's plan table shows document classification and extraction only on its Standard and Enterprise plans, which are quoted. Its Document Understanding docs then charge each page in Platform Units, UiPath's own usage unit. UiPath's pages don't give a dollar price for a unit.

Nanonets charges per workflow step, from $0.02 for simple steps to $0.30 for data extraction, and says a typical invoice runs 4 to 6 steps.

Hyperscience's own site shows no price and says its cost per page falls as yearly volume grows. Its AWS Marketplace listing, sold by Hyperscience, prices a 12-month cloud contract at $50,000 and doesn't say how many pages that covers.

The cost no price list includes#

Some fields come back wrong or unsure, and a person has to check them. Microsoft suggests flagging low-confidence results for human review. Amazon's review service, A2I, takes no new customers, and Google deprecated Document AI's Human-in-the-Loop feature on January 16, 2024. On the clouds, your team builds the review step and staffs it.

Review time can cost more than the software. Put your own numbers in the calculator below. For the share of documents reviewed, use what a person had to check or fix in a pilot on your own files.

The calculator starts at 5,000 documents a month, 4 pages each, at $0.01 a page. Software comes to $200 a month. If a person spends 4 minutes on one document in ten, at $35 an hour, review adds $1,167 a month. The total is $0.27 a document, and 85% of it is review time.

Now try a tool that costs three times as much per page but sends only 3% of documents to a person. Software rises to $600 a month and review falls to $350. The total drops to $950 a month, or $0.19 a document. To size the savings side of the budget, see how to measure IDP ROI.

Our take. We'd judge each quote by what a finished document costs on your own files. A vendor's accuracy figure can't tell you how many of your documents will need a person. Run a few hundred real documents through each tool on your shortlist, and count the ones someone had to fix. A document that someone had to fix costs more than the price list says, because you also pay for their time.

Questions that make two quotes comparable#

You can compare a page rate with a quote only when you know what each one covers. Ask every vendor these questions, and get the answers in writing.

Where Docsumo fits#

Docsumo is an IDP platform your team works in, with an API for your own systems. Its pricing page lists three plans. The Free plan is a 14-day trial with up to 1,000 pages and 10 user licenses, set up with you on a short call. Business and Enterprise are custom quotes, and you can pay monthly, or annually for a lower rate.

The plans differ by the steps they add. Business adds auto-classification and splitting, a test environment and audit logging. Enterprise adds cross-document validations, master data lookup and automated workflows. Docsumo gives each field a confidence score, a number for how sure it is. Fields that score below the threshold you set for that field go to your own reviewers. Docsumo provides the software, not a review team.

Docsumo is cloud only, so it can't run on your own servers. Software companies that build Docsumo into their product use the document extraction API. They start with the free trial and agree terms before launch. We compare Docsumo with each cloud service on its own page, for Amazon Textract, Azure Document Intelligence and Google Document AI. The pricing page lists every plan feature, and the Enterprise page covers security reviews and hosting regions.

Price one month of your own files before you sign#

Take a month of real documents, including the messy ones. Ask each vendor on your shortlist to price that exact volume. Then run a pilot, count what people had to fix, and put both numbers in the calculator. The tool with the lowest cost per finished document is the cheapest one, whatever its page price.

Book a demo and bring that month of files, or start a free trial.

Frequently asked questions#