Last updated August 2026. Google sells two different OCR products and they are priced an order of magnitude apart. Cloud Vision API text detection costs $1.50 per 1,000 pages after a free first 1,000 each month. Document AI Form Parser, the one that actually returns structured tables and key-value pairs, costs $30.00 per 1,000 pages. Picking the wrong one is a 20x billing mistake, and the pricing pages that would tell you this sit on separate URLs.
Every figure below was read from Google's own published pricing on cloud.google.com in August 2026, not from a summary. Prices are US dollars and, unless stated, do not vary by region for these products.
Google Cloud Vision API pricing
Vision API bills per image. For a PDF, every page counts as one image, and every feature you apply to that page is a separate billable unit. Run text detection and label detection over the same page and you pay for two units, not one. The first 1,000 units of each feature each month are free.
| Feature | First 1,000 / month | Units 1,001 to 5,000,000 | Above 5,000,000 |
|---|---|---|---|
| Text Detection | Free | $1.50 per 1,000 | $0.60 per 1,000 |
| Document Text Detection | Free | $1.50 per 1,000 | $0.60 per 1,000 |
| Label Detection | Free | $1.50 per 1,000 | $1.00 per 1,000 |
| Object Localization | Free | $2.25 per 1,000 | $1.50 per 1,000 |
| Web Detection | Free | $3.50 per 1,000 | Contact Google |
| Logo, Landmark, Facial Detection | Free | $1.50 per 1,000 | $0.60 per 1,000 |
| Image Properties | Free | $1.50 per 1,000 | $0.60 per 1,000 |
| Crop Hints | Free | Free with Image Properties, or $1.50 | Free with Image Properties, or $0.60 |
| Safe Search Detection | Free | Free with Label Detection, or $1.50 | Free with Label Detection, or $0.60 |
Two details in that table cost people money. Crop Hints and Safe Search are free only when bundled with the feature they are paired with, and billed at full rate otherwise. And the final 1,000-unit block is prorated: Google bills in blocks of 1,000, so 4,300 requests is charged as four full blocks plus 300/1000 of a fifth, not as 4.3 blocks rounded down.
Google Document AI pricing
Document AI is the product to use when you need structure rather than a wall of text. It understands that a table is a table and that an invoice has a total. That understanding is what you are paying 20 times more for.
| Processor | Volume tier | Price per 1,000 pages |
|---|---|---|
| Enterprise Document OCR | First 1,000 pages / month | Free |
| Enterprise Document OCR | 1,000 to 5,000,000 | $1.50 |
| Enterprise Document OCR | Above 5,000,000 | $0.60 |
| OCR add-on capabilities | All volumes | $6.00 |
| Layout Parser (includes initial chunking) | All volumes | $10.00 |
| Form Parser | Up to 1,000,000 | $30.00 |
| Form Parser | Above 1,000,000 | $20.00 |
| Custom extractor | Up to 1,000,000 | $30.00 |
| Custom extractor | Above 1,000,000 | $20.00 |
| Re-chunking parsed documents | All volumes | $0.02 |
Specialized parsers are priced per document rather than per page, which changes the arithmetic completely for multi-page files:
| Specialized parser | Price |
|---|---|
| Invoice parser | $0.10 per document |
| Expense parser (formerly receipt parser) | $0.10 per document |
| Bank statement parser | $0.75 per classified document |
| Pay slip parser | $0.30 per classified document |
| W2 parser | $0.30 per classified document |
| Lending document splitter and classifier | $0.05 per classified document |
| US driver license parser | $0.10 per document |
| US passport parser | $0.10 per document |
| Identity document proofing | $0.10 per document |
There is also a hosting charge that catches teams who deploy a custom processor and leave it running: $0.05 per hour, or $36.50 per month, for a deployed processor version, and $300.00 per month for provisioned throughput. That is a fixed cost that accrues whether or not you send a single page through it.
How Google compares to AWS and Azure on the same job
All three hyperscalers land on the same number for plain OCR, which is more interesting than it looks: $1.50 per 1,000 pages is the going rate for reading text off a page, everywhere. The divergence is entirely in structured extraction, and it is large.
| Job | AWS Textract | Azure Document Intelligence | |
|---|---|---|---|
| Plain OCR, per 1,000 pages | $1.50 | $1.50 | $1.50 |
| OCR at high volume | $0.60 above 5M | $0.60 above 1M | Commitment tiers |
| Table extraction, per 1,000 pages | $30.00 (Form Parser) | $15.00 (Tables) | $10.00 (Layout) |
| Forms and key-value pairs | $30.00 (Form Parser) | $50.00 (Forms) | $10.00 (prebuilt) |
| Custom trained extraction | $30.00 | $25.00 (Custom Queries) | $30.00 |
| Document layout | $10.00 (Layout Parser) | $4.00 (Layout) | $10.00 (Layout) |
| Free tier | 1,000 pages / month | None ongoing | Free tier available |
For table extraction specifically, Azure Layout at $10.00 per 1,000 pages is a third of Google Form Parser at $30.00. If your workload is genuinely just "get the tables out of these PDFs" and you have no other reason to be on Google Cloud, that gap is the whole decision. Where Google wins is the ongoing free tier: 1,000 pages a month, every month, on both Vision and Document AI OCR, which AWS does not match beyond its initial trial period.
What this actually costs at real volumes
Rates per 1,000 pages are hard to feel. Here is the monthly bill for one job, converting scanned PDFs into structured spreadsheet data, at four volumes on Google.
| Pages per month | Vision text detection | Document AI OCR | Document AI Form Parser |
|---|---|---|---|
| 1,000 | $0.00 | $0.00 | $30.00 |
| 10,000 | $13.50 | $13.50 | $300.00 |
| 100,000 | $148.50 | $148.50 | $3,000.00 |
| 1,000,000 | $1,498.50 | $1,498.50 | $30,000.00 |
The Vision and Document AI OCR columns are identical because the rate and the free allowance are identical. The Form Parser column is the one that surprises people: at 100,000 pages a month, choosing structured extraction over plain OCR is a $2,850 monthly difference. That is the real question to answer before you write any code, and it is a question about your documents, not about Google.
Which Google product should you use?
Use Cloud Vision text detection when you need the words and you will do the structuring yourself, or when the document is simple enough that reading order is sufficient. Use Document AI Enterprise Document OCR at the same price when you want OCR with better document awareness and the add-on capabilities available. Use Layout Parser at $10.00 when you need document structure, headings and chunks, which is the common choice for retrieval pipelines. Use Form Parser at $30.00 only when you genuinely need tables and key-value pairs returned as structured data and you have tested that a cheaper path fails.
That last test matters more than the pricing table. A large share of business PDFs are not scans at all: they have a real text layer, and the tables in them can be reconstructed by reading the text positions directly, at zero OCR cost. Running a born-digital PDF through Form Parser at $30 per 1,000 pages is paying for machine vision on a document that never needed it. Check whether your files have a text layer before you price an OCR pipeline, because for many finance and reporting workflows the honest answer is that the conversion accuracy problem is not an OCR problem.
Frequently asked questions
Is the Google Cloud Vision API free?
Partly. The first 1,000 units of each feature are free every month, permanently, not just during a trial. Beyond that you pay $1.50 per 1,000 units for text detection. New Google Cloud accounts also receive general free credits that can cover early usage. For a low-volume workload of under 1,000 pages a month, Vision API genuinely costs nothing on an ongoing basis.
How much does Google Document AI cost per page?
It depends entirely on the processor. Enterprise Document OCR is $0.0015 per page, or $1.50 per 1,000. Layout Parser is $0.01 per page. Form Parser and custom extractors are $0.03 per page, dropping to $0.02 above one million pages a month. Specialized parsers such as the invoice parser are billed per document at $0.10 rather than per page.
What is the difference between Cloud Vision and Document AI?
Vision API is general image analysis that happens to read text. Document AI is built for documents specifically and understands document structure: pages, tables, form fields, entities. For plain text extraction they cost the same, $1.50 per 1,000 pages, so the choice is about output shape rather than price. Once you need tables or fields returned as data, only Document AI does it, at 20 times the OCR rate.
Does Google charge per page or per document?
Both, depending on the processor. General processors such as OCR, Layout Parser and Form Parser bill per page, and a 10-page PDF is 10 billable pages. Specialized parsers including the invoice, expense, W2 and bank statement parsers bill per document regardless of page count, which makes them dramatically cheaper for long multi-page files in their specific categories.
Is Document AI cheaper than AWS Textract?
For plain OCR they are identical at $1.50 per 1,000 pages. For tables, AWS Textract is cheaper at $15.00 per 1,000 against Google's $30.00. For forms and key-value pairs, Google is cheaper at $30.00 against Textract's $50.00. Neither is universally cheaper, so price the specific processors your pipeline will call rather than comparing the vendors in the abstract. Our full breakdown of Textract pricing has the complete AWS rate card.
Are there hidden costs in Google Document AI?
Two worth budgeting for. Deployed custom processor versions carry a hosting charge of $0.05 per hour, which is $36.50 a month whether or not you use them, and provisioned throughput is $300.00 a month. On the Vision side, every feature applied to an image is billed separately, so a request that asks for text detection and label detection on the same page costs two units. Storage and network egress are billed by Google Cloud separately in both cases.
Do you get billed for failed requests?
No. Google states that requests returning 4xx or 5xx response codes are not billed on these APIs. That matters when you are developing against the API and sending malformed requests, and it means a retry loop on transient failures does not multiply your bill for the failures themselves, only for the eventual successful call.
Where a per-page API stops being the right tool
Cloud OCR pricing rewards scale. At a million pages a month, engineering a pipeline around a $1.50 rate is obviously worth the build cost. At two hundred statements a month it is not: you will spend more on the developer time to build and maintain the pipeline than the API will ever bill you, and you still have to solve the part the API does not do, which is turning raw extraction output into a spreadsheet somebody can actually use.
That break-even sits lower than most teams expect. Before committing to a cloud OCR build, price the alternative honestly, including the ongoing maintenance of a service whose output format changes when the vendor updates a model. For teams whose real requirement is a finished spreadsheet rather than a JSON response, a tool priced per file rather than per page is usually both cheaper and finished sooner.
The per-document parsers make the same point from the other direction. Google charges $0.75 for a bank statement and $0.10 for a receipt because it already knows what those documents contain, and that knowledge is most of the value. If the job behind your receipt parsing is really categorizing employee spend and closing the month, the parser is one component of a workflow you would otherwise be building around it. Price the workflow, not the API call.