All three hyperscalers list basic OCR at $1.50 per 1,000 pages. Their volume tiers start in different places, and one of them does not discount until five million pages a month. Here is the real curve from 100,000 to 10 million pages, with the commitment tiers that are not on the pricing pages.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Uploading...
Run a real document from your pipeline and check the field accuracy before you model the volume.
At high volume, basic OCR runs between $0.45 and $1.50 per 1,000 pages depending on vendor and contract, and structured field extraction runs between $18 and $50. The spread inside that range is decided almost entirely by where each vendor's volume tier begins, not by the list price, because AWS Textract Detect Document Text, Azure Read and Google Enterprise Document OCR all list at exactly the same $1.50 per 1,000 pages.
AWS and Azure begin their discounted tier above one million pages a month. Google does not begin its discounted tier until above five million. That single difference means a company running two million pages a month pays about $2,100 on AWS, about $1,200 on an Azure commitment tier, and about $3,000 on Google, for output that is broadly comparable. The list-price comparison that put them level was accurate and useless.
Last updated July 2026. Google's tier boundaries were read off Google's own pricing page and the Azure commitment tiers off the Azure Retail Prices API, which is the feed the portal bills from, both on July 23, 2026.
Text extraction only, on the meter each vendor positions for it: Textract Detect Document Text, Azure Read, and Google Enterprise Document OCR. The three are level until a million pages, then they separate.
| Pages per month | AWS Textract | Azure Read, pay as you go | Azure, best commitment tier | Google Enterprise Document OCR |
|---|---|---|---|---|
| 100,000 pages | $150 | $150 | $150 | $150 |
| 500,000 pages | $750 | $750 | $375 | $750 |
| 1,000,000 pages | $1,500 | $1,500 | $750 | $1,500 |
| 2,000,000 pages | $2,100 | $2,100 | $1,200 | $3,000 |
| 5,000,000 pages | $3,900 | $3,900 | $3,000 | $7,500 |
| 10,000,000 pages | $6,900 | $6,900 | $5,260 | $10,500 |
How to read this. The AWS and Azure pay-as-you-go columns apply the published graduated tiers, so a 2 million page month bills the first million at $1.50 and the second at $0.60. The Azure commitment column is our arithmetic over Microsoft's published commitment and overage rates, picking the cheapest combination at each volume, and it is our calculation rather than a figure Microsoft quotes as a package. The Google column stays at $1.50 all the way to five million pages because that is where its first tier ends. Negotiated enterprise pricing exists on every vendor here and none of them publishes it, so treat this curve as the floor of the conversation.
Google's pricing page sets Enterprise Document OCR tier one at 1 to 5,000,000 pages a month at $1.50 per 1,000, and tier two at 5,000,001 and above at $0.60. AWS Textract and Azure Document Intelligence both put the same $0.60 rate above one million pages. Every one of those four numbers is public and none of them is surprising in isolation. Put side by side, they mean that the band between one and five million pages a month, which is where a great many real document pipelines actually sit, is the band where the three vendors are furthest apart despite listing identically.
At two million pages a month the gap is $900 against AWS. At five million it is $3,600, and Google is still on its opening rate while the others have been discounted for four million pages. Only above five million does Google's rate finally match, and by then the cumulative difference for the year is substantial. Notably, Google's structured meters do not behave this way: Form Parser and Custom Extractor both tier at 1,000,001 pages, the same boundary the other two use. It is specifically the basic OCR meter where the boundary sits five times further out.
This is not an argument that Google is the wrong choice. Its per-request page ceiling, its parser lineup and the fact that it does not bill failed requests all matter, and we cover those on the OCR pricing per 1,000 pages reference. It is an argument that a list-price comparison is the wrong tool once you are past a million pages, and that the tier boundary belongs in your model before the vendor logo does.
Read off the Azure Retail Prices API for East US on July 23, 2026. This is the feed the portal bills from, and it exposes tiers the marketing pricing page does not show.
| Commitment tier | Monthly fee | Overage rate | Effective rate at the allowance |
|---|---|---|---|
| Read 500K | $375 / mo | $0.75 / 1k | $0.75 per 1,000 |
| Read 2M | $1,200 / mo | $0.60 / 1k | $0.60 per 1,000 |
| Read 8M | $4,200 / mo | $0.53 / 1k | $0.53 per 1,000 |
| Read 16M | $7,200 / mo | $0.45 / 1k | $0.45 per 1,000 |
| Prebuilt 500K | $4,000 / mo | $8.00 / 1k | $8.00 per 1,000 |
| Prebuilt 1M | $7,500 / mo | $7.50 / 1k | $7.50 per 1,000 |
| Custom 500K | $10,500 / mo | $21.00 / 1k | $21.00 per 1,000 |
| Custom 1M | $18,000 / mo | $18.00 / 1k | $18.00 per 1,000 |
Two things stand out. The 500,000 page Read commitment reaches $0.75 per 1,000 pages at half the volume where the pay-as-you-go discount even begins, so a company running half a million pages a month is paying twice what it needs to if it never asked. And the 16 million tier at $0.45 is 25 percent below the best pay-as-you-go rate anyone can reach on any of the three clouds. The connected container tiers run lower still, 15 to 20 percent below the equivalent Azure tier, though those bill through a deployment you host and operate yourself.
Four of these fourteen meters are flat at every volume. If your pipeline is built on one of them, growth does not bend your cost curve at all.
| Meter | Vendor | List rate | At volume |
|---|---|---|---|
| Detect Document Text | AWS Textract | $1.50 / 1k | $0.60 above 1M pages/mo |
| Read | Azure AI Document Intelligence | $1.50 / 1k | $0.60 above 1M, or $0.45 on the 16M commitment |
| Enterprise Document OCR | Google Document AI | $1.50 / 1k | $0.60, but only above 5M pages/mo |
| Analyze Expense | AWS Textract | $10 / 1k | $8 above 1M pages/mo |
| Tables | AWS Textract | $15 / 1k | $10 above 1M pages/mo |
| Forms | AWS Textract | $50 / 1k | $40 above 1M pages/mo |
| Analyze ID | AWS Textract | $25 / 1k | $10 after the first 100K pages |
| Queries | AWS Textract | $15 / 1k | No discount at any volume |
| Custom extraction | Azure AI Document Intelligence | $30 / 1k | $20 above 1M, or $18 on the 1M commitment |
| Custom Extractor | Google Document AI | $30 / 1k | $20 above 1M pages/mo |
| Form Parser | Google Document AI | $30 / 1k | $20 above 1M pages/mo |
| Layout Parser | Google Document AI | $10 / 1k | No discount at any volume |
| Documents, standard output | Amazon Bedrock Data Automation | $10 / 1k | No discount at any volume |
| Documents, custom output | Amazon Bedrock Data Automation | $40 / 1k | No discount at any volume |
Amazon Bedrock Data Automation is the outlier worth naming. It has no volume discount on any meter at any scale, and its custom output rate of $40 per 1,000 pages is the most expensive flat custom rate among the big three. It also charges an extra $0.0005 per field per page above 30 fields, so a wide schema raises the rate rather than leaving it flat. We break that down on the Amazon Bedrock Data Automation pricing reference.
All three hyperscalers list basic OCR at $1.50 per 1,000 pages. Between one and five million pages a month, two of them have already halved and one has not. Comparing list prices at that volume gives you exactly the wrong answer.
Google Layout Parser and AWS Textract Queries are single tier at every volume, and no Bedrock Data Automation meter discounts at any scale. If your pipeline is built on one of those, your cost curve is a straight line and no amount of growth bends it.
Volume tiers count per billing account and usually per region. Two accounts each running 600,000 pages a month both sit in the expensive band, while one account running 1.2 million would not. Consolidate before you model.
Azure publishes commitment tiers through its Retail Prices API and its enterprise documentation, not on the marketing pricing page most people read. The 500,000 page commitment reaches $0.75 per 1,000 at half the volume where the pay-as-you-go discount even begins.
At these volumes the binding constraint is often pages per minute rather than dollars per page. Google caps an online request at 15 pages, Azure takes 2,000 in a single call, and provisioned capacity on Google costs $300 per extra page-per-minute per month on top of the per-page rate.
Batch is where the discount lives on the language model readers, and it is also named in the zero data retention exclusion lists. If your contract depends on that arrangement, the discounted path may not be available to you.
Once volume is genuinely large, the alternative to a per-page meter is a GPU you rent by the hour. The arithmetic turns on utilization rather than on the model being free. A per-page API bills nothing when no documents arrive; a GPU instance bills all 730 hours in a month whether it is busy or idle, so the real rate is the monthly instance cost divided by the pages you actually processed.
On the cheapest current AWS card suited to the work, that break-even against the $1.50 per 1,000 cloud rate lands somewhere around 390,000 pages a month. Below it the GPU is dramatically more expensive, and at 10,000 pages a month it is roughly 39 times the cloud rate. Above it the GPU pulls ahead and keeps going. What makes the decision hard is that the break-even assumes sustained utilization, and most real document volume is spiky. We worked the whole calculation through, including the hidden costs, on self hosted OCR cost.
There is also a constraint that arrives before the money does. At these volumes the ceiling is often pages per minute rather than dollars per page, and the per-request limits differ enormously: Google caps an online request at 15 pages while Azure accepts 2,000 in a single call. Those ceilings, and what it costs to raise them, are on the OCR API limits comparison.
DocuOCR plans work out to roughly $14 to $20 per 1,000 pages across the published tiers, and we are not going to pretend that is cheaper than $0.60 for raw text extraction, because it is not and it is not the same product. What you get for the difference is classification of a mixed batch, named field extraction with per-field confidence, validation rules, a review queue for the values that fail them, and an export that lands in your system rather than a JSON blob you still have to reconcile.
Here is the honest test. If what you need is text off a page and you already have engineers who will build classification, field mapping, validation and human review around it, a hyperscaler meter at volume is the cheaper path and you should take it. If what you need is extracted, checked data and you would otherwise be building that layer yourself, compare our rate against the meter plus the engineering, because the meter is rarely the expensive part. For very large or steady volume, talk to us before you model off the published tiers, which is the same advice this page gives about every vendor on it.
Run a real document from your pipeline through DocuOCR, check the fields and the confidence scores, and see what the extraction layer is worth before you price the meter underneath it.
OCR pricing per 1,000 pages
Every vendor list rate in one table
OCR API pricing comparison
The buyer view across the market
Batch OCR software
Processing thousands of files in one run
Self hosted OCR cost
Where a GPU finally beats the meter
OCR API limits comparison
The throughput ceilings that bite first
OCR API data retention
What each vendor keeps, and for how long
From the same family of tools
