Adobe prices PDF Services by the Document Transaction, and PDF Extract consumes one transaction for every five pages, rounded up. The Free Tier includes 500 Document Transactions per month. Adobe publishes that unit in full detail and publishes no dollar rate at all: paid usage is quoted by sales under a separate written agreement.
So the useful number is transactions, not dollars. This page converts your document shape into monthly Document Transactions so you can price any quote Adobe gives you. Last updated August 2026.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Free plan extracts the first 5, rest can be unlocked after
Uploading...
Extract returns the document's structure. Drop in the file you were going to parse and see what finished, named fields look like instead.
Adobe bills PDF Services API by the Document Transaction, which its licensing page defines as being "based on the initial endpoint request (i.e., API call) and the digital output". PDF Extract and PDF To Markdown consume one transaction for every five pages. Almost every other operation, including OCR, Export, Create and Combine, gets fifty pages per transaction. Accessibility Auto-Tag runs the other way entirely and charges ten transactions for each individual page. You get 500 transactions a month free, and past that Adobe quotes you privately.
That last part is the whole story. We downloaded Adobe's pricing page and its pay-as-you-go page in full on August 5, 2026, about 1.4 MB of markup each, and swept both for currency amounts. Neither contains a single dollar figure. A page titled "pay as you go" that does not say what you pay is unusual, and it means Adobe cannot be compared with AWS, Azure, Google or Mistral on published numbers. It can only be compared on the unit, which is what the rest of this page does.
Transcribed from Adobe's published metrics table on August 5, 2026. This is the part Adobe commits to in writing, and it is the only part you can plan against before you talk to sales.
| Operation | What it does | Document Transactions |
|---|---|---|
| Extract PDF and PDF To Markdown | Returns text, tables, figures and reading order as JSON, or LLM-friendly Markdown | 1 per 5 pages |
| OCR | Converts a scanned PDF into a searchable, selectable PDF | 1 per 50 pages |
| Export | Converts PDF to Word, Excel, PowerPoint, RTF, JPEG or PNG | 1 per 50 pages |
| Create | Builds a PDF from HTML, Office files or images | 1 per 50 pages |
| Combine | Merges multiple PDFs into one | 1 per 50 pages |
| Compress, Linearize, Split | Reduces size, optimizes for web view, or splits into smaller PDFs | 1 per 50 pages |
| Insert, Replace, Delete, Rotate, Reorder Pages | Page-level edits, counted on the resulting output document | 1 per 50 pages |
| Document Generation | Builds a PDF or Word file from a Word template plus JSON data | 1 per document |
| Accessibility Auto-Tag | Tags PDF content so assistive technology can read it | 10 per page |
| Electronic Seal | Applies an organizational seal using a digital certificate | 10 per PDF |
| Protect, Remove Password, Get PDF Properties | Password and metadata operations | 1 per document, no page count |
Two details in that table catch people out. Page-editing operations are counted on the output document, not the input, which is why Adobe's own example of inserting ten pages into a 100-page PDF costs three transactions rather than two. And Document Generation is billed per document regardless of length, so a template that produces a fifteen-page PDF still costs one transaction.
Nobody publishes this table, so here it is. Multiply the transaction count by whatever rate Adobe quotes you and you have a per-1,000-pages figure you can put next to AWS, Azure and Google.
| Operation | Transactions per 1,000 pages, packed | Transactions for 1,000 one-page documents | Penalty for short documents |
|---|---|---|---|
| Accessibility Auto-Tag | 10,000 | 10,000 | Charged per page, so packing changes nothing |
| Extract PDF and PDF To Markdown | 200 | 1,000 | 5x penalty for one-page documents |
| OCR, Export, Create, Combine, page edits | 20 | 1,000 | 50x penalty for one-page documents |
Read the first column top to bottom. Accessibility Auto-Tag costs 10,000 transactions per 1,000 pages and OCR costs 20. That is a 500x spread between two operations on the same API key, and it is the single most expensive thing to get wrong when you architect against Adobe. Auto-Tag is a genuinely heavy piece of work, so the ratio is not unreasonable, but it is invisible unless you read the metrics table carefully.
The right-hand columns are the rounding trap. Adobe states that transaction counts are "rounded up on a 5-page basis for Extract PDF and PDF To Markdown and on a 50-page basis for other PDF Tools", and its own example confirms the consequence: converting ten single-page Word documents consumes ten transactions. If you push 1,000 one-page invoices through OCR you pay 1,000 transactions rather than the 20 the same page count would cost packed into fifty-page files. This is the same shape as Google's 10-page billing block on its Invoice Parser, and it bites the same buyers for the same reason: real business documents are short.
Comparisons of Adobe's operations usually divide fifty by five and declare Extract ten times more expensive. That is only true at the top of the range. Because both meters round up per document, the premium is a function of document length, and below six pages it is exactly zero.
| Your document | Extract transactions | OCR transactions | Extract premium |
|---|---|---|---|
| 1-page invoice or receipt | 1 | 1 | None |
| 5-page contract | 1 | 1 | None |
| 6-page contract | 2 | 1 | 2x |
| 20-page bank statement | 4 | 1 | 4x |
| 50-page report | 10 | 1 | 10x |
| 400-page manual (the Extract page ceiling) | 80 | 8 | 10x |
For an accounts payable team running one-page invoices, or a lender pulling data off short forms, Extract and OCR cost the same number of transactions, and the choice comes down purely to whether you want structured JSON or a searchable PDF. The crossover is six pages. Everything above that is where the premium starts, and it caps at ten times once a document reaches fifty pages. If most of your volume is long reports, that ten times is real and worth designing around.
Adobe's Free Tier credentials process 500 Document Transactions per month with no credit card. The same 500 transactions is worth anywhere between 50 pages and 25,000 pages depending only on which endpoint you call.
| If you use it for | 500 transactions reaches |
|---|---|
| OCR on 50-page documents | 25,000 pages a month |
| Extract on 5-page documents | 2,500 pages a month |
| Extract or OCR on single-page documents | 500 documents a month |
| Electronic Seal | 50 documents a month |
| Accessibility Auto-Tag | 50 pages a month |
Note the scanned-document ceiling. A 150-page scan is the practical limit for Extract, PDF To Markdown and Auto-Tag, well below the 400-page limit for born-digital PDFs, because scanned pages carry an OCR pass.
| Product | Public rate? | What you can plan against today |
|---|---|---|
| Adobe PDF Extract API | No public rate | Unit published (1 transaction per 5 pages), rate quoted by sales |
| AWS Textract Detect Document Text | Published | About $1.50 per 1,000 pages |
| Azure AI Document Intelligence Read | Published | About $1.50 per 1,000 pages |
| Google Enterprise Document OCR | Published | About $1.50 per 1,000 pages |
| Mistral OCR 4 | Published | About $4 per 1,000 pages |
| DocuOCR | Published | Published plans, about $14 to $20 per 1,000 pages |
Search for Adobe PDF Extract pricing and you will run into a per-call figure of about five cents, repeated across listicles and forum answers. It traces to Adobe's old AWS Marketplace listing. Adobe Community Manager Raymond Camden announced on May 10, 2023 that the API "will no longer be offered on the AWS Marketplace for new subscribers as we've decided to sunset the listing", and that announcement quotes no price at all.
So the number circulating today is a community figure attached to a listing that has been closed to new customers for over three years. We are not going to reprint it as a rate. If a vendor comparison quotes you a current Adobe per-page price, ask where it came from.
Extract returns the structure of a PDF as JSON: text runs with their bounding boxes and font details, tables, figures, headings and reading order. It is a genuinely strong parser, and PDF To Markdown packages the same understanding as Markdown for feeding a language model. If your problem is that a PDF's layout destroys naive text extraction, this is the class of tool that fixes it.
It does not hand you named business fields. There is no invoice_number, no total_due, no vendor_tax_id. Getting from a table of text runs to a validated record is your work: classification to decide what each document is, field mapping, rules that catch a total which does not match its line items, a screen where a person fixes what the model was unsure about, and an export into your accounting system.
That is a scope boundary rather than a flaw, and it is the same one we draw for open models that return Markdown rather than fields. It matters for budgeting because the transaction count is only part of the cost. For most US mid-market teams, the engineering wrapped around a parsing API costs more in year one than the API calls do. If you want to compare the finished-pipeline route against a per-transaction one, the OCR API pricing comparison puts the published rates side by side, and PDF table extraction covers the specific case of getting clean rows out of a document.
Azure, AWS Textract and Google Document AI per 1,000 pages, side by side.
A cross-vendor rate reference for every meter we have verified.
Per-feature rates from Detect Document Text through Analyze Lending.
S0 rates for Read, Layout, prebuilt and custom extraction.
Including the $438 a year idle processor hosting fee.
Why the billing unit changes the answer more than the rate does.
Getting clean rows out of a document, not just text.
Turning a PDF into structured records your systems can read.
Page ceilings, file size caps and rate limits across the major APIs.
A transaction count cannot tell you whether the fields come out right. Upload one of your real documents, look at what comes back, and then decide whether you are buying a parser or a finished pipeline.
