Google Document AI bills the invoice and expense parsers at 0.10 dollars per count, where one count is up to ten pages, so a single page receipt costs ten cents no matter how short it is. Enterprise Document OCR is 1.50 dollars per 1,000 pages, and Form Parser and custom extractors are 30 dollars per 1,000. Every figure here was read off the Google Cloud Document AI pricing page in August 2026. Upload a receipt below to see the extracted fields without a Google Cloud project.
Upload your receipts and invoices
Drop files here or click to upload
Upload failed, please try again
Up to 50 files
Uploading...
Google prices its OCR products in three different units on two different pages. Some processors bill per 1,000 pages, some bill per count of up to ten pages, and Cloud Vision bills per feature applied to an image. Quote the wrong unit and your estimate is off by a factor of ten, which is exactly what happens in most of the pricing summaries that circulate online.
The invoice, expense, utility and identity parsers are priced at 0.10 dollars per count, and Google defines one count as up to ten pages in a document. A one page receipt and a ten page invoice cost the same ten cents. Most write-ups convert that to a flat 0.01 dollars per page, which is only true if every document you send is exactly ten pages.
Enterprise Document OCR gives you text and coordinates at 1.50 dollars per 1,000 pages. The expense parser gives you merchant, date, tax and total as named fields at 0.10 dollars per count. Teams costing a receipt pipeline routinely price it on the OCR meter and then discover they need the parser meter.
Cloud Vision has its own pricing page and its own billing unit. Every feature you apply to an image is a separate billable unit, so running text detection and label detection on one image bills two units. Vision has no receipt or invoice parser at all, which is the reason most people end up on Document AI.
A deployed custom processor version costs 0.05 dollars per hour whether or not you send it anything. Google prices one deployed version for a year at 438 dollars. That is a standing charge nobody includes in a per page estimate.
These are the published US rates from the Google Cloud Document AI pricing page in August 2026. Watch the unit column carefully: the pretrained parsers are billed per count of up to ten pages, while the OCR, form and custom processors are billed per 1,000 pages.
Formerly the receipt parser. Returns merchant, date, line items, tax and total. One count covers up to ten pages, so a single receipt still costs ten cents. This is the meter that matters for receipt work.
Same rate and same ten page count rule. A twelve page invoice is two counts, so twenty cents. Google states that 1 to 10 pages costs 0.10, 11 to 20 costs 0.20, and 91 to 100 costs 1.00 dollar.
The first 1,000 counts each month are free. From 1,000 to 5 million it is 1.50 dollars per 1,000, and above 5 million it drops to 0.60. Text and coordinates only, with no named fields.
Key value pairs and tables from documents you have no trained model for. Falls to 20 dollars per 1,000 above a million pages. Layout Parser, including initial chunking, is 10 dollars per 1,000.
Models you train on your own documents, dropping to 20 dollars per 1,000 above a million pages. Custom splitters and classifiers are 5 dollars per 1,000, falling to 3 dollars at volume.
Charged for every hour a custom processor version stays deployed, independent of traffic. Google prices one version for a year at 438 dollars. The provisioned service tier is a separate 0.410958904 dollars per hour.
The count rounding is what makes or breaks a Google estimate, so start there rather than with a page total.
Google rounds up to the next ten page block per document. Split your monthly volume into 1 to 10 page documents, 11 to 20 page documents, and so on, then multiply each group by the number of ten page blocks it needs.
Tip: A pile of single page receipts is the worst case for this billing model. Ten separate receipts cost 1.00 dollar. One ten page invoice costs 0.10.
If you need named fields, you are on the expense or invoice parser at 0.10 dollars per count. If you only need the raw text, Enterprise Document OCR at 1.50 dollars per 1,000 pages is roughly 66 times cheaper for a one page document. Choose before you estimate.
Add 0.05 dollars per hour for every deployed custom processor version, the Cloud Storage the async API requires for batch jobs, and the engineering weeks to wire the JSON into your accounting system. For a small finance team that build cost dwarfs the inference charges.
Every rate below is published by Google on cloud.google.com. Nothing here is estimated, averaged, or copied from a third party summary.
You need the real per count and per 1,000 page meters before you can decide between building on Document AI and buying a finished extraction tool.
You have Document AI line items on an invoice and need to know which processor is driving them. The pretrained parsers at 0.10 dollars per count are usually the answer.
The ten page count rule works against you. A thousand one page receipts cost 100 dollars on the expense parser, the same as ten thousand pages bundled into thousand page documents.
If you want receipts as an Excel file and have no interest in running a Google Cloud project, the per count rate is not the number that decides this. The build effort is.
Google splits Document AI into two price lists that use different units. Pretrained parsers such as invoice and expense are billed per count, where Google defines one count as up to ten pages in a document. Everything else, including OCR, Form Parser and custom extractors, is billed per 1,000 pages. Getting the unit right is most of the work.
| Processor | Price | Unit | What it returns |
|---|---|---|---|
| Expense parser (formerly receipt parser) | $0.10 | per count, up to 10 pages | Merchant, date, line items, tax, total |
| Invoice parser | $0.10 | per count, up to 10 pages | Invoice header and line item fields |
| Utility parser | $0.10 | per count, up to 10 pages | Utility bill fields |
| US driver license and passport parsers | $0.10 | per count | Identity document fields |
| Bank statement parser | $0.75 | per classified document | Statement header and transactions |
| Pay slip parser | $0.30 | per classified document | Payroll fields |
| W-2 parser | $0.30 | per classified document | W-2 boxes |
| Procurement and lending splitter or classifier | $0.05 | per classified document | Document type and split points |
| Enterprise Document OCR | Free to 1,000, then $1.50, then $0.60 above 5M | per 1,000 count | Text and coordinates, no named fields |
| OCR add-ons | $6.00 | per 1,000 | Optional extras on top of OCR |
| Layout Parser (includes initial chunking) | $10.00 | per 1,000 | Structure and chunks |
| Form Parser | $30.00, then $20.00 above 1M | per 1,000 | Key value pairs and tables |
| Custom extractor | $30.00, then $20.00 above 1M | per 1,000 | Fields you trained |
| Custom splitter or classifier | $5.00, then $3.00 above 1M | per 1,000 | Type label and split points |
| Summarizer | $0.025 | per document | Generated summary |
| Re-chunking parsed documents | $0.02 | per 1,000 count | New chunk boundaries |
| Custom processor hosting | $0.05 | per hour deployed | Standing charge, no traffic required |
This is the single most misreported fact about Google Document AI pricing. The pretrained parsers are not billed per page. They are billed per count, and Google states plainly that one count equals up to ten pages in a document. The published examples are unambiguous: a document of 1 to 10 pages costs 0.10 dollars, a document of 11 to 20 pages costs 0.20 dollars, and a document of 91 to 100 pages costs 1.00 dollar. Rounding is per document and always upward.
Plenty of comparison posts translate the parser rate into a flat 0.01 dollars per page. That figure is only correct when every document you send happens to contain exactly ten pages. For the workload most businesses actually have, a pile of one page receipts, the real rate is 0.10 dollars per page, ten times higher.
| Document length | Counts billed | Cost on the expense parser | Effective cost per page |
|---|---|---|---|
| 1 page receipt | 1 | $0.10 | $0.100 |
| 3 page invoice | 1 | $0.10 | $0.033 |
| 10 page invoice | 1 | $0.10 | $0.010 |
| 11 page invoice | 2 | $0.20 | $0.018 |
| 20 page invoice | 2 | $0.20 | $0.010 |
| 100 page statement pack | 10 | $1.00 | $0.010 |
The practical consequence: batching matters more on Google than on any other OCR vendor. If your source documents genuinely belong together, merging ten single page receipts from one vendor into one ten page PDF turns 1.00 dollar of parsing into 0.10 dollars. If they do not belong together, do not merge them, because you will get one set of fields back rather than ten.
All three clouds sell a structured receipt or invoice extractor, and the headline rates look similar until you apply the unit. AWS Textract AnalyzeExpense is 0.0100 dollars per page. The Azure prebuilt receipt and invoice models are 10 dollars per 1,000 pages, which is the same 0.01 dollars per page. Google is 0.10 dollars per count of up to ten pages. On a one page receipt Google is ten times the price of either. At exactly ten pages the three are level, and Google never comes out cheaper.
| Vendor and model | Rate | 1 page receipt | 10 page invoice |
|---|---|---|---|
| Google Document AI expense parser | $0.10 per count, 10 pages per count | $0.100 | $0.100 |
| AWS Textract AnalyzeExpense | $0.0100 per page | $0.010 | $0.100 |
| Azure prebuilt receipt or invoice | $10.00 per 1,000 pages | $0.010 | $0.100 |
| Google Enterprise Document OCR (text only) | $1.50 per 1,000 | $0.0015 | $0.015 |
| AWS Textract DetectDocumentText (text only) | $0.0015 per page | $0.0015 | $0.015 |
| Azure Read OCR (text only) | $1.50 per 1,000 pages | $0.0015 | $0.015 |
Raw OCR is priced identically across the three clouds at 1.50 dollars per 1,000 pages, which tells you the commodity has no pricing power left in it. The difference is entirely in the structured extraction layer. Our OCR API pricing comparison puts all the vendor meters side by side, and the Azure Document Intelligence pricing reference has the full Azure meter list including commitment tiers.
Cloud Vision is the other place people look for Google OCR pricing, and it is a separate product on a separate price list with a different billing unit: one unit is one feature applied to one image, so two features on the same image bills twice. Its text detection rate matches Document AI OCR, but it has no receipt, invoice or expense parser at all, so a receipt project that starts there ends up writing its own field parsing. The full rate card is in our Google Cloud Vision API pricing breakdown, and if you have already hit that wall, the Google Vision OCR alternative page covers the switch.
Take a bookkeeping practice processing 2,500 single page receipts a month. On the expense parser that is 2,500 counts at 0.10 dollars, so 250 dollars a month in inference alone. The identical workload on AWS Textract AnalyzeExpense is 25 dollars, and on the Azure prebuilt receipt model it is also 25 dollars. Nothing about Google is worse technically here. The count rounding simply penalizes short documents, and receipts are the shortest documents there are.
Flip the workload to 250 ten page invoice packets and the picture changes completely: Google charges 25 dollars, AWS and Azure charge 25 dollars, and the three are level. If your documents are long, Google costs what everyone else costs. If they are single page receipts, budget ten times as much or use a different processor.
Google wins on document variety and on chunking for retrieval work. The specialized parsers cover territory the other clouds do not sell as prebuilt models, including pay slips at 0.30 dollars per document, W-2s at 0.30, and bank statements at 0.75. Layout Parser at 10 dollars per 1,000 with chunking built in is aimed squarely at document ingestion for retrieval augmented generation, and re-chunking already parsed documents at 0.02 dollars per 1,000 is cheap enough to iterate on. If you are building a document pipeline inside Google Cloud and your documents are long and varied, Document AI is a reasonable buy. If you are a finance team with a folder of receipts, the arithmetic above is the whole argument.
| Google Document AI | ReceiptOCR | |
|---|---|---|
| What you get back | JSON entities with confidence values | Excel, CSV, or a QuickBooks ready file |
| Setup required | GCP project, billing account, service credentials, SDK | Upload in a browser |
| Billing unit | Per count of up to 10 pages, rounded up | Flat monthly plan with a page allowance |
| Single page receipt cost | $0.10 on the expense parser | Included in the plan |
| Standing charges | $0.05 per hour per deployed custom processor | None |
| Best for | Engineering teams with varied, long documents | Finance teams who need the spreadsheet |
These are different purchases. Google sells inference and leaves the pipeline to you. If you have already decided Document AI is more machinery than the job needs, the Google Document AI alternative page covers the switch, and receipt OCR software covers what a finished tool does instead. For why none of these vendors publish a receipt accuracy figure, see OCR accuracy.
Last updated August 2026. Google Cloud prices change and vary by currency and region, so confirm on cloud.google.com before you commit to a volume.
The invoice, expense and utility parsers cost 0.10 dollars per count, where one count is up to ten pages in a document. Enterprise Document OCR is 1.50 dollars per 1,000 pages after a free first 1,000, dropping to 0.60 above five million. Form Parser and custom extractors are 30 dollars per 1,000 pages, falling to 20 dollars above a million.
Enterprise Document OCR works out at 0.0015 dollars per page, or 1.50 dollars per 1,000, with the first 1,000 free each month and a 0.60 dollar rate above five million. That meter returns text and coordinates only. Structured receipt or invoice fields come from the parser meter instead, at 0.10 dollars per count.
For the pretrained parsers, one count equals up to ten pages in a single document, rounded up. The Google published examples state that a 1 to 10 page document costs 0.10 dollars, an 11 to 20 page document costs 0.20 dollars, and a 91 to 100 page document costs 1.00 dollar. Rounding happens per document, so ten separate one page receipts are ten counts.
The invoice parser is 0.10 dollars per count, the same rate as the expense parser, with one count covering up to ten pages. A three page invoice is one count at 0.10 dollars. A twelve page invoice is two counts at 0.20 dollars. There is no separate per field or per line item charge.
Only partly. Enterprise Document OCR gives you the first 1,000 counts each month at no charge, and Cloud Vision gives you the first 1,000 units per feature each month free. The invoice and expense parsers have no free allowance, so structured receipt extraction is billed from the first document.
Form Parser is 30 dollars per 1,000 pages, dropping to 20 dollars per 1,000 above a million pages a month. It returns key value pairs and tables from documents you have no trained model for. Layout Parser, which includes initial chunking, is cheaper at 10 dollars per 1,000, and custom extractors match Form Parser at 30 dollars.
Not for short documents. AWS Textract AnalyzeExpense is 0.0100 dollars per page and the Azure prebuilt receipt model is 10 dollars per 1,000 pages, both of which work out at a cent for a one page receipt. Google charges 0.10 dollars for that same receipt. At exactly ten pages per document the three are level, and Google is never cheaper.
Yes. Custom processor hosting is 0.05 dollars per hour for every deployed version, billed whether or not you send it any documents. The published Google example puts one processor version deployed for a year at 438 dollars. Budget it as a fixed monthly line, not a variable one.
Stop typing receipts by hand
Upload your receipts and invoices and get a clean Excel or CSV file in minutes.
Extract my receipts nowFree to try, no sign up required
Google, AWS, Azure, Veryfi and Mindee rates compared per page.
Every Azure OCR meter and commitment tier, verified August 2026.
Receipt and invoice data to Excel without a Google Cloud project.
Why Vision returns words and boxes rather than receipt fields.
Why no major vendor publishes a receipt extraction accuracy rate.
