Azure OCR runs through the Read feature of Azure AI Vision. On the S1 tier it costs $1.50 per 1,000 transactions for the first million a month and $0.60 per 1,000 after that, and every PDF page is one transaction. A free F0 tier covers 5,000 transactions a month. Read returns lines and words, not invoice fields. Upload an invoice below to see the vendor, totals and every line item already structured.
PDF, JPG, PNG, BMP, HEIC and TIFF
Drop files here or click to upload
Up to 50 files
Uploading...
Azure Read is one of the cheapest ways to turn a scanned page into text. The mistake is treating $1.50 per 1,000 pages as the price of getting vendor invoices into your books.
The Read API gives you pages, lines, words, bounding boxes and confidence scores. It does not say which number is the invoice total, which date is the due date or where the line-item table starts.
Microsoft counts each page of a multipage PDF as a separate transaction. A 200-page document is 200 transactions, so page count, not file count, drives the bill.
The same Read model is sold in Azure AI Vision and in Azure Document Intelligence. Microsoft points document work to the Document Intelligence version, so teams often price the wrong product first.
The older OCR feature bills at the cheaper Group 1 rate of $1.00 per 1,000, but Microsoft says it does not recommend that API and it receives no further updates. Building on it to save 50 cents is a false economy.
When teams find that Read cannot label invoice fields, the Azure answer is the Document Intelligence prebuilt invoice model at $10 per 1,000 pages, or Content Understanding field extraction at $22 in West US.
Upload handling, polling the asynchronous Read operation, parsing text into columns, vendor matching, review and the export to QuickBooks, NetSuite or Excel all sit outside the API.
These rates come from the Azure Retail Prices API for the East US region, cross-checked against Microsoft's Azure AI Vision pricing page and the Vision OCR documentation. Prices are in USD per 1,000 transactions.
The free tier allows 5,000 transactions a month at 20 per minute. Free tier Read processes only the first two pages of a PDF and caps images at 4 MB, so it suits testing, not production.
Read sits in Image Analysis Group 2. On the S1 tier it costs $1.50 per 1,000 transactions for the first 1 million a month and $0.60 per 1,000 above 1 million.
Read accepts PDF and TIFF files up to 2,000 pages. Each page counts as a transaction, so a 3-page invoice costs three.
Read can be prepaid in monthly blocks: 500,000 transactions for $375, 2 million for $1,200, 8 million for $4,200 and 16 million for $7,200, with overage at each tier's effective rate.
The Read container on a connected commitment runs $338, $1,080, $3,780 and $6,480 a month for the same four tiers. Disconnected containers are billed yearly at the same annual total.
Tagging, objects, brands, color and the legacy OCR API cost $1.00 per 1,000 for the first million, stepping down to $0.40 above 100 million. Captions and dense captions bill with Read in Group 2.
The meter is simple once you count pages. The honest number also needs the work Microsoft does not bill you for.
Add up the pages you will send in a month. A mid-sized AP team at 3,000 invoices averaging 1.4 pages sends about 4,200 pages, which is $6.30 a month on S1.
Tip: The F0 tier is a separate resource, not a free allowance on top of S1.
Below about 250,000 pages a month pay as you go wins. At 500,000 pages the commitment tier costs $375 instead of $750, and at 2 million it costs $1,200 instead of $2,100.
Budget either a field model (Document Intelligence prebuilt invoice, Content Understanding, an LLM pass) or the developer time to parse the text yourself. This decides whether Read is actually the cheap option.
A general OCR endpoint and an invoice extraction tool are priced for different jobs. Here is the honest split.
Search indexing, archive digitization, reading labels in photos or feeding page text into your own model are exactly what Read is priced for.
If identity, storage and billing already live in your Azure tenant, the setup cost is small and Read slots into existing Logic Apps or Functions pipelines.
If the goal is vendor, date, totals and line items in a spreadsheet, raw OCR is only the first step. A finished extractor costs less than the developer hours needed to finish the job.
Provisioning a Vision resource, managing keys, polling async results and writing a parser is the wrong project for a finance team that needs bills entered this week.
Azure OCR is sold as the Read feature of Azure AI Vision, the service Microsoft used to call Computer Vision. On the standard S1 tier, Read costs $1.50 per 1,000 transactions for the first 1 million transactions in a month and $0.60 per 1,000 after that. A transaction is one image or one page of a PDF or TIFF. There is no subscription fee on S1 and no minimum. The free F0 tier gives 5,000 transactions a month at 20 per minute.
| Azure AI Vision meter (S1, East US) | Up to 1M / month | 1M to 10M | 10M to 100M | Above 100M |
|---|---|---|---|---|
| Group 2: Read (OCR), Caption, Dense Captions, Describe | $1.50 | $0.60 | $0.60 | $0.60 |
| Group 1: Tag, Objects, Brands, Color, Smart Crops, legacy OCR | $1.00 | $0.65 | $0.60 | $0.40 |
| Custom image classification or object detection | $2.00 per 1,000 predictions, training $20 per hour | |||
| Free F0 tier | 5,000 transactions a month, 20 per minute, first 2 PDF pages only, 4 MB image cap | |||
Prices are per 1,000 transactions in USD, read from the Azure Retail Prices API for East US. Microsoft's own pricing page lists the same feature groups. The volume bands are monthly and reset on the first of the month.
Commitment tiers apply only to Read. You prepay a monthly block of transactions and any overage bills at roughly the block's effective rate. The same blocks are sold for the Read container, either connected (it reports usage to Azure) or disconnected (fully offline, billed yearly).
| Read block | Azure-hosted / month | Effective per 1,000 | Connected container / month | Disconnected container / year | Same volume pay as you go |
|---|---|---|---|---|---|
| 500,000 | $375 | $0.75 | $338 | Not offered | $750 |
| 2,000,000 | $1,200 | $0.60 | $1,080 | $12,960 | $2,100 |
| 8,000,000 | $4,200 | $0.525 | $3,780 | $45,360 | $5,700 |
| 16,000,000 | $7,200 | $0.45 | $6,480 | $77,760 | $10,500 |
Three details matter here. The 500,000 block halves the pay-as-you-go price, which is the biggest discount on the card, and it pays for itself once you pass 250,000 pages a month. The connected container is exactly 10% cheaper than the hosted block at every size. And the disconnected container costs precisely twelve months of the connected price, so Microsoft charges no premium for running Read fully offline. That matters to US government contractors and healthcare teams who cannot send documents to a public endpoint.
On S1, Azure OCR costs $0.0015 per page for the first million pages in a month and $0.0006 per page after that. A 3-page invoice costs $0.0045. With a 2 million commitment the effective price is $0.0006 per page from the first page, and with the 16 million block it drops to $0.00045.
For an AP department the numbers are small. 3,000 invoices a month at 1.4 pages each is 4,200 pages, or $6.30. Even 50,000 invoices a month is roughly $105. OCR is almost never the expensive part of an invoice project on Azure.
They use the same OCR engine and cost the same $1.50 per 1,000 pages, but Microsoft positions them for different inputs. Its OCR documentation recommends Azure AI Vision Read (version 4.0) for general, in-the-wild images such as labels, signs and product photos, and the Document Intelligence Read model for scanned and digital documents such as reports, forms and invoices. Document Intelligence adds an asynchronous API tuned for text-heavy pages and sits next to the prebuilt invoice model on the same resource.
Microsoft has also stopped developing the older Vision v3.2 Read API. Its documentation says all future Read enhancements go into the two services above. If you are starting an invoice project today, price Document Intelligence, not Vision. We break those meters down, including the prebuilt invoice model and its commitment tiers, in Azure Document Intelligence pricing.
Read gives you text. To get a vendor name, invoice number, due date, total and line items as fields, Azure sells two field products, and both cost several times the OCR rate.
| Option | Cost per 1,000 invoice pages | Returns invoice fields? | What you still build |
|---|---|---|---|
| Azure AI Vision Read | $1.50 | No, lines and words only | All field and line-item parsing, export |
| Document Intelligence Read | $1.50 | No, text and layout | All field and line-item parsing, export |
| Document Intelligence prebuilt invoice | $10.00 | Yes | JSON mapping, review, export |
| Content Understanding field extraction | $22.00 (West US only in the US) | Yes | Analyzer schema, review, export |
| AWS Textract Analyze Expense | $10.00 | Yes | JSON mapping, review, export |
| Google Document AI Invoice Parser | $100 on single-page invoices ($0.10 per document) | Yes | JSON mapping, review, export |
| InvoiceExtractor | Flat plan, from $49 a month for 2,500 pages | Yes, header and every line item | Nothing, export Excel, CSV or JSON |
The prebuilt invoice model costs 6.7 times the Read rate, and Content Understanding field extraction costs almost 15 times. That is still cheap per page. The expensive part is everything after the JSON arrives: mapping fields to your chart of accounts, stitching line items that break across pages, a review screen for low-confidence values, and the export into QuickBooks, NetSuite or Excel. For most finance teams that engineering outweighs a year of API charges. Our Azure Content Understanding pricing page covers the newer field meters in detail.
For plain text the three clouds charge the same list price. Azure Read, AWS Textract Detect Document Text and Google Cloud Vision Document Text Detection all cost $1.50 per 1,000 pages at normal volume and $0.60 at high volume. Azure and AWS both drop to $0.60 after 1 million pages a month, while Google waits until 5 million. Azure also sells fixed prepaid Read blocks, which AWS does not offer for Textract, so at 500,000 pages a month Azure costs $375 against $750 on Textract pay as you go.
The difference that matters for invoices is the field model. Textract Analyze Expense and the Azure prebuilt invoice model are both $10 per 1,000 pages, while Google bills its Invoice Parser per 10-page document block. See Amazon Textract pricing and Google Cloud Vision API pricing for the full rate cards.
Read is the right buy when text is the product. Full-text search over scanned archives, reading labels and signs in photos, feeding page text into your own model or an Azure OpenAI pipeline, or adding OCR to an app already running in your Azure tenant all fit the per-transaction model well. At $1.50 per 1,000 pages, or $0.45 with the largest block, it is hard to beat.
It is the wrong buy when the requirement is "get these vendor invoices into the ledger with every line item by month end." Then the OCR bill is the smallest number in the project. A finished invoice extractor reads the header and each line, lets a person check the result, and exports Excel, CSV or JSON at a flat monthly price with no Azure resource, key or polling loop. Upload one of your own invoices in the tool above and compare the output with a Read response before you decide which one your team should maintain.
Azure OCR (the Read feature of Azure AI Vision) costs $1.50 per 1,000 transactions for the first 1 million a month on the S1 tier and $0.60 per 1,000 after that. Each image or PDF page is one transaction. Commitment blocks start at $375 a month for 500,000 transactions.
Azure AI Vision has a free F0 tier with 5,000 transactions a month at 20 per minute. It only reads the first two pages of a PDF and caps images at 4 MB, so it suits testing. F0 is a separate resource tier, not a free allowance on top of paid S1 usage.
Read, Caption and Dense Captions cost $0.0015 per image for the first million a month. Group 1 features such as tagging, object detection and brands cost $0.001 per image. Each feature you request on an image is billed as its own transaction.
Both sell the Read OCR model at $1.50 per 1,000 pages. Document Intelligence adds document models such as the prebuilt invoice model at $10 per 1,000 pages and layout at $10. Microsoft recommends Document Intelligence Read for scanned and digital documents and Vision Read for general images.
Yes, for Read only. Monthly blocks cost $375 for 500,000 transactions, $1,200 for 2 million, $4,200 for 8 million and $7,200 for 16 million. Connected containers cost 10% less, and disconnected containers are billed yearly at the same annual total.
Read extracts text, not invoice fields. It does not label the vendor, invoice number, total or line items. For fields, Azure sells the Document Intelligence prebuilt invoice model at $10 per 1,000 pages and Content Understanding field extraction at $22 in West US.
On pay as you go they match, $1.50 per 1,000 pages dropping to $0.60 after 1 million. Azure sells prepaid Read blocks and AWS does not, so at 500,000 pages a month Azure costs $375 against $750 on Textract.
Yes. InvoiceExtractor reads invoice headers and every line item and exports Excel, CSV or JSON with no Azure subscription, resource or API key. Plans start at $49 a month, or $24 a month billed yearly, for 2,500 pages.
Azure Document Intelligence pricing for Read, the prebuilt invoice model and commitment tiers.
Azure Content Understanding pricing, field extraction at $22 per 1,000 pages.
Amazon Textract pricing, Analyze Expense at $10 per 1,000 invoice pages.
Google Cloud Vision API pricing for OCR at $1.50 per 1,000 units.
AWS Textract vs Azure Document Intelligence for invoice extraction.
Invoice OCR software that returns fields and line items, not raw text.
Invoice data extraction API that returns invoice fields as JSON.
Start turning your invoices into clean, structured spreadsheet data.
USD
per month
per month
$288 charged today
Choose speed vs accuracy when extracting
| Base AI Faster | 2,500 pages |
| Pro AI Best accuracy | 500 pages |
Scale invoice extraction across your whole team with automation.
USD
per month
per month
$888 charged today
Choose speed vs accuracy when extracting
| Base AI Faster | 10,000 pages |
| Pro AI Best accuracy | 2,000 pages |
Enterprise‑grade invoice extraction, security, and controls.
USD
per month
per month
$ charged today
Choose speed vs accuracy when extracting
| Base AI Faster | pages |
| Pro AI Best accuracy | pages |
Secure checkout powered by Stripe