Google Cloud Vision charges about $1.50 per 1,000 units for text detection and document text detection, after a free allowance of 1,000 units a month. Above 5,000,000 units a month the rate falls to about $0.60 per 1,000. One unit is one feature on one image, and each page of a PDF counts as one image. Every figure here comes from Google's own published pricing page.
Written for US teams costing out an OCR vendor. Google revises rates and varies them by region, so confirm on the current Cloud Vision pricing page. Last updated July 2026.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Uploading...
Before you wire up a Vision pipeline, drop in the document you meant to test on it and see what finished, validated fields look like.
Cloud Vision is cheap, and the pricing is simple once you know what a unit is. The first 1,000 units each month are free. From 1,001 to 5,000,000 units a month, text detection and document text detection cost about $1.50 per 1,000. Past 5,000,000 units they drop to about $0.60 per 1,000. A unit is one feature applied to one image, so asking for two features on the same image bills twice, and every page of a PDF is its own image. Run 35,000 pages a month through document text detection and you will pay roughly $51. The catch is not the price. It is what you get for it: Cloud Vision hands back text and coordinates, never named fields, never table structure, never a trained document type. Teams who choose it for invoices end up writing the extraction layer themselves.
Published rates as of July 2026, taken from Google's own Cloud Vision pricing page. Tiers are cumulative within a single calendar month and reset each month. Final blocks of 1,000 units are prorated, so a partial block is not billed as a whole one.
| Feature | First 1,000 units a month | 1,001 to 5,000,000 units | 5,000,001 units and above |
|---|---|---|---|
| Text detection (TEXT_DETECTION) | Free | About $1.50 per 1,000 | About $0.60 per 1,000 |
| Document text detection (dense text, PDFs) | Free | About $1.50 per 1,000 | About $0.60 per 1,000 |
| Label detection | Free | About $1.50 per 1,000 | About $1.00 per 1,000 |
| Key-value form fields | Not offered on Cloud Vision | Not offered on Cloud Vision | Not offered on Cloud Vision |
| Table structure | Not offered on Cloud Vision | Not offered on Cloud Vision | Not offered on Cloud Vision |
| Custom trained document types | Not offered on Cloud Vision | Not offered on Cloud Vision | Not offered on Cloud Vision |
Two rows in that table matter more than the prices. Label detection does not fall to $0.60 at high volume the way text detection does, it stops at about $1.00, which surprises teams who assume one discount applies to everything. And the bottom three rows are all the same answer: Cloud Vision does not do structured extraction at any price. For a cross-vendor view of what those jobs cost elsewhere, see our OCR API pricing comparison.
Almost every Cloud Vision billing surprise traces back to one sentence in Google's documentation: each feature applied to each image is a separate billable unit. Read that literally, because the billing system does.
Send one image and request DOCUMENT_TEXT_DETECTION and LABEL_DETECTION in the same call and you are billed two units. Four features on one image is four units. The single API request is a convenience, not a discount.
Each page of a PDF or TIFF counts as one image. A 40-page contract is 40 units under one feature, about $0.06. Multi-page files go through the asynchronous endpoint, but the per-page arithmetic is identical.
Google prorates the final partial block of 1,000 units, so processing 34,200 units does not bill as 35,000. You pay for what you sent, rounded at the block boundary rather than up to it.
Take a realistic US mid-market workload: 35,000 pages a month of mixed business documents. Here is what Cloud Vision bills, and what it still leaves you holding.
$51
per month
34,000 billable units after the free 1,000, at about $1.50 per 1,000. You receive text and bounding boxes.
$103
per month
70,000 units, because each feature on each page bills separately. Still no named fields.
$1,050
per month
About $30 per 1,000 pages, and it does return key-value pairs. The right Google product for forms.
$499
per month
The published 35,000 page plan, about $14 per 1,000 pages, with classification, validation, review, and export included.
Compare the first column to the last and Cloud Vision looks ten times cheaper. That comparison is only fair if raw text is the finished product. If your actual goal is an invoice number, a vendor name, a total, and a date landing in your accounting system, the $51 buys you the starting line. Someone still has to write the parsing rules, decide what to do when the total sits in an unexpected place, build the screen where a person checks the low-confidence values, and keep all of it working as vendors redesign their invoices. That work does not appear on a Google bill, which is exactly why it gets left out of comparisons.
Google sells two things that read text, and the pricing pages do not make the difference obvious. Cloud Vision is an image API. Document AI is a document API. Picking the cheap one for a document job is the most expensive mistake in this table.
| What you need | Google Cloud Vision | Google Document AI | DocuOCR |
|---|---|---|---|
| Plain text off a page | About $1.50 per 1,000 units | About $1.50 per 1,000 pages (Enterprise Document OCR) | Included in plan |
| Invoices and receipts as named fields | Not offered, you build the parsing | About $10 per 1,000 pages (prebuilt parsers, billed in 10-page blocks) | Included in plan |
| Any form, as key-value pairs | Not offered | About $30 per 1,000 pages (Form Parser) | Included in plan |
| Your own document type, trained | Not offered | About $30 per 1,000 pages (Custom Extractor) | Included, no training needed |
| Document classification | Not offered | About $5 per 1,000 pages (Custom Classifier) | Included in plan |
| Idle hosting fee | None | About $0.05 an hour per deployed custom processor version | None |
| Free allowance | 1,000 units a month | No standing free tier, trial credit only | Test on your own files, no signup |
| Human review of low-confidence values | You build the screen | You build the screen | Included review screen |
If you are reading signage, product photos, screenshots, or scanned images where the text itself is the answer, Cloud Vision is the correct and cheapest choice, and you should stop reading here and go use it. If you are reading invoices, bank statements, insurance forms, contracts, or anything with a layout that means something, you want Google Document AI pricing instead, or a product that already includes the pipeline around it.
Vision returns text and coordinates. Deciding that the number to the right of the word "Total" is the total, on every vendor template you receive, is code you write and own.
Someone has to look at the values the model was unsure about. That screen, the queue behind it, and the audit trail are all yours to build.
A mixed batch of invoices, statements, and contracts has to be sorted before it can be parsed. Vision does not classify documents.
The API call is one line item. Cloud Storage, the compute that orchestrates it, and egress charges arrive separately on the same bill.
For most US mid-market teams the engineering around the API costs more in year one than the API does, and it recurs every year after in maintenance. That is not an argument against Cloud Vision. It is an argument for pricing the whole job rather than the API call, which is the same advice we would give about AWS Textract pricing or Azure Document Intelligence pricing.
Azure, AWS Textract and Google Document AI per 1,000 pages, side by side.
The other Google product, including the $438 a year idle processor hosting fee.
Per-feature rates from Detect Document Text through Analyze Lending.
S0 rates for Read, Layout, prebuilt and custom extraction.
What OCR 4 actually costs per 1,000 pages, and why the web still quotes the old rate.
A capability head-to-head rather than a price sheet.
Which one fits which document workload.
What a document extraction API returns when the pipeline is already built.
An honest roundup of the options for US teams.
A rate card 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 an OCR call or a finished pipeline.
