Google Cloud Vision API Pricing: OCR Cost

Last updated August 2026.

The Google Cloud Vision API charges 1.50 dollars per 1,000 units for text detection and document text detection, after a free allowance of 1,000 units per feature per month. Above five million units a month the rate drops to 0.60 dollars per 1,000. A unit is one feature applied to one image, so a single image sent with two features enabled is billed twice. Vision has no receipt or invoice parser, so the price buys you words and bounding boxes, not named fields.

Google Cloud Vision API pricing table (US list, August 2026)

Google publishes one price list for the whole Vision API and prices each feature separately. The first 1,000 units of each feature are free every month, and the final 1,000 unit block of a month is prorated rather than rounded up.

How is a Vision API unit calculated?

A unit is one feature applied to one image, not one API request and not one image. Google gives the example directly: apply face detection and label detection to the same image and you are billed for one unit of label detection and one unit of face detection. Two features on one photo is two units, and the free tier is counted per feature, so you get 1,000 free label detections and a separate 1,000 free text detections in the same month.

This is the detail that breaks most Vision cost estimates. Teams count images, multiply by 1.50 dollars per 1,000, and then find the bill is double or triple because the request had several features switched on. Before you model anything, open the actual request payload and count the features in it.

How much does Google Vision OCR cost per image?

Text detection works out at 0.0015 dollars per image, which is 1.50 dollars per 1,000. Above five million units a month it falls to 0.0006 dollars. Document text detection, the variant tuned for dense pages rather than photos of signs, is priced identically. Neither rate includes any charge for the number of words on the page, so a receipt and a full page of contract text cost the same.

Is the Google Vision API free?

There is a permanent free allowance rather than a trial: the first 1,000 units of each feature every month cost nothing, and that resets monthly. For a low volume side project that can genuinely be free forever. It is not a free tier in the sense of a capped sandbox, and there is no free allowance on Object Localization or Web Detection beyond the same 1,000 units. New Google Cloud accounts also get promotional credit, which is separate from the Vision allowance.

Can the Vision API read PDFs?

Yes, with three conditions. The supported MIME types are application/pdf and image/tiff only. The files have to live in a Cloud Storage bucket, because Google documents that PDF and TIFF document detection is only available for files stored in Cloud Storage buckets. And the hard ceiling is 2,000 pages: the Vision API accepts PDF and TIFF files up to 2,000 pages, and larger files return an error. PDF and TIFF text detection is billed at the document text detection or text detection rate.

In practice that means a PDF workflow on Vision is an asynchronous one. You upload to a bucket, kick off a long running operation, poll it, and read JSON back out of another bucket. That is more moving parts than a single synchronous call, and it is the point where a lot of small teams decide the API is not the product they wanted.

Does the Vision API extract receipt fields?

No. Vision returns text with bounding boxes and a hierarchy of pages, blocks, paragraphs, words and symbols. It has no concept of a merchant, a purchase date, a sales tax line or a grand total. Everything that turns raw text into an accounting record, finding the total among a dozen dollar figures, deciding which line is the vendor name, splitting tax out of the subtotal, is code you write and maintain.

That is the single most important thing to understand before pricing a receipt project on Vision. The API charge is small. The parsing layer is the project. Google sells the parsing layer separately under Document AI, where the expense parser returns named fields at 0.10 dollars per count, and the full rate card is on our Google Document AI pricing reference.

Which is cheaper for OCR, Cloud Vision or Document AI?

For raw text they cost the same. Vision text detection and Document AI Enterprise Document OCR are both 1.50 dollars per 1,000, both drop to 0.60 dollars above five million, and both give you a free first 1,000. For structured receipt or invoice fields there is no comparison to make, because Vision does not offer them at any price.

The honest summary: Vision is the cheapest way to get characters off an image inside Google Cloud, and it is the wrong tool if what you actually need is a spreadsheet row per receipt. Our OCR API pricing comparison puts Google, AWS, Azure and the specialist receipt vendors on the same per page basis, and the Azure Document Intelligence pricing page covers the closest Microsoft equivalent.

What a Vision based receipt pipeline really costs

Price out 2,500 receipts a month, the volume a small bookkeeping practice handles. Vision text detection on 2,500 images with one feature is 1,500 billable units after the free tier, so 2.25 dollars. That number is real and it is also close to meaningless, because it excludes the Cloud Storage bucket, the service account and IAM setup, the retry logic, the parsing rules that find the total, and the person who fixes those rules every time a vendor changes its receipt layout.

The failure mode worth planning for is silent degradation. OCR does not usually break loudly. It starts returning slightly fewer blocks per image, or splits a total across two words, and the job still exits zero. A pipeline that quietly returns worse data looks identical to a pipeline that is working, which is why teams running extraction at volume tend to monitor row counts and schema drift on what lands downstream rather than trust an exit code. On a receipt workflow the equivalent check is arithmetic: line items plus tax should equal the stated total, and a mismatch is a free signal that something was misread.

When Vision is the right buy, and when it is not

Vision earns its place when you need text off images at scale and the downstream consumer is search, moderation, translation or a language model rather than an accounting ledger. It handles more than 200 languages, the rate is as low as this market goes, and the free monthly allowance is genuinely useful for prototypes. Object localization, logo detection and safe search sit on the same API, so a mixed image workload is cheaper to run in one place.

It is the wrong buy when the deliverable is a finance record. If the job is turning a folder of receipts and invoices into rows in Excel or a QuickBooks import, the API rate is not the number that decides anything, and a finished tool skips the entire build. That is the argument laid out on our Google Vision OCR alternative page, and receipt OCR software covers what the finished category does differently. If you are still comparing engines on accuracy claims, read OCR accuracy first, because no major vendor publishes a receipt extraction accuracy rate at all.

Three things to check before you commit to Vision pricing

First, count features rather than images, because the unit is the feature and that is where estimates double. Second, confirm your documents are images rather than multi page PDFs, because PDF work forces you into Cloud Storage and asynchronous operations. Third, decide whether you are buying text or buying fields. If the answer is fields, Vision is not on the shortlist and the comparison you actually want is Document AI against AWS Textract against a finished extraction tool.

Prices here are the published US list rates from the Google Cloud Vision pricing page as of August 2026. Google changes cloud pricing and quotes different figures in other currencies, so confirm on cloud.google.com before you sign off a volume commitment.