Azure Document Intelligence costs $1.50 per 1,000 pages to read plain text, $10.00 for Layout and every prebuilt model, and $30.00 for custom extraction. Those rates are the same in every commercial Azure region on earth. The bill only really moves when you change deployment model, and that is what this page prices out.
Written for US teams costing out a document extraction vendor. Every figure was computed from Microsoft's public Azure Retail Prices API across all 38 regions, with the meter name and effective date shown. Last updated August 2026.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Free plan extracts the first 5 files, the rest unlock when you upgrade
Uploading...
Before you commit to a tier, drop in the document you meant to test on Azure and see what finished, validated fields look like.
Reading a page of plain text costs $1.50 per 1,000 pages, dropping to $0.60 above a million pages a month. Asking for structure costs a lot more: Layout and every prebuilt model (invoice, receipt, ID, W-2) are $10.00 per 1,000 pages, and a model you trained yourself is $30.00. Classification is $3.00, add-on features are $6.00 and query fields are $10.00, both charged on top of whatever model you called.
The part that surprises people is what does not change the price. Region does not. Batch does not. Going over a commitment tier does not. The three levers that genuinely move an Azure extraction bill are which model you call, how many pages you send it, and whether you run it in Microsoft's cloud or in a container you host.
Where these numbers come from. Microsoft's own pricing page renders every figure as a dash until you select a region and currency, which makes it hard to cite and impossible to diff. The numbers on this page were pulled instead from the public Azure Retail Prices API at prices.azure.com, which needs no account, and were computed across all 38 regions that carry Document Intelligence meters. Each rate below shows the exact meter name Microsoft bills it under and the date that rate took effect.
Pay-as-you-go (S0) rates in US East, which as the next section shows is the same as every other commercial region. The meter name column is what you will actually see on the Azure invoice, and it is the string to search for when a line item does not look familiar.
| What you are calling | Model ID | Azure meter name | Per 1,000 pages | Volume break | Rate effective |
|---|---|---|---|---|---|
| Read (plain text OCR) | prebuilt-read | S0 Read Pages | $1.50 | $0.60 above 1M pages a month | 2022-03-01 |
| Layout and every prebuilt model | prebuilt-layout, prebuilt-invoice, prebuilt-receipt, prebuilt-idDocument, prebuilt-tax.us.w2 | S0 Pre-built Pages | $10.00 | No published volume break | 2021-08-01 |
| Document classification | Custom classifier | S0 pages for doc classifier | $3.00 | No published volume break | 2023-09-01 |
| Custom extraction | Your trained model | S0 Custom Pages | $30.00 | No published volume break | 2024-06-01 |
| Custom generative extraction | Your generative model | S0 Custom Generative Pages | $30.00 | No published volume break | 2024-06-01 |
| Add-ons (High Resolution, Font, Formula) | Request feature flags | S0 Add-on for Pages | $6.00 | Charged on top of the model | 2024-11-01 |
| Query fields | queryFields | S0 pages for query fields | $10.00 | Charged on top of the model | 2024-11-01 |
| Custom model training | Training run | S0 Training | $3.00 per hour | Billed by the hour, not the page | 2024-11-01 |
One naming note worth keeping in mind when you read your invoice: in the retail feed the service family is now called Foundry Tools, while the product is still Azure Document Intelligence and older resources may still say Form Recognizer. If your bill uses the old name, the Form Recognizer pricing breakdown covers what changed and what did not.
No, and this is worth knowing because the answer is different on AWS. We pulled every Document Intelligence meter in all 38 regions Microsoft publishes. Read is $1.50 in 31 of the 33 regions that carry it. Prebuilt is $10.00 in 32 of 34. Custom is $30.00 in 32 of 34. The exceptions are not regions at all in the ordinary sense: they are the two Azure Government regions, US Gov Arizona and US Gov Virginia, and they charge exactly 1.25 times the commercial rate.
| Meter | Every commercial region | Azure Government | Multiplier |
|---|---|---|---|
| Read (S0 Read Pages) | $1.50 | $1.875 | 1.25x |
| Layout and prebuilt models | $10.00 | $12.50 | 1.25x |
| Custom extraction | $30.00 | $37.50 | 1.25x |
| Add-on features | $6.00 | $7.50 | 1.25x |
| Document classification | $3.00 | $3.75 | 1.25x |
| Batch Layout | $10.00 | $12.50 in US Gov Virginia, $12.60 in US Gov Arizona | 1.25x and 1.26x |
Textract and Azure Read carry the same $1.50 list price for plain text, so the two look interchangeable on a spreadsheet. They are not, once region enters the picture. Amazon's own price list feed puts US West (N. California) at 1.40 times the standard rate on every Textract meter, and GovCloud at 1.26 times. Deploying into us-west-1 instead of us-west-2 costs 40 percent more for byte-identical output. On Azure that mistake does not exist, because there is only one commercial price. The full Textract breakdown is in our AWS Textract pricing analysis.
Every Azure Government meter is exactly 1.25 times its commercial twin, with a single exception. Batch Layout is $12.50 per 1,000 pages in US Gov Virginia and $12.60 in US Gov Arizona, a 1.26 multiplier where everything else is 1.25. It is ten cents per thousand pages and it changes nobody's decision, but it is in the live feed and it is the kind of thing Microsoft tends to correct quietly, so treat it as accurate for August 2026 and re-check before you build a model on it.
Containers are not billed per page. You buy a monthly commitment tier, and the same tier costs less in a connected container than it does against the cloud endpoint. The discount is published but it is never stated as a percentage, so almost nobody notices that it is different for each model family. It is exact across all twelve tiers.
| Model | Commitment tier | Azure cloud, per month | Connected container, per month | You save |
|---|---|---|---|---|
| Read | 500,000 pages a month | $375 | $338 | 10% |
| Read | 2,000,000 pages a month | $1,200 | $1,080 | 10% |
| Read | 8,000,000 pages a month | $4,200 | $3,780 | 10% |
| Read | 16,000,000 pages a month | $7,200 | $6,480 | 10% |
| Custom extraction | 20,000 pages a month | $540 | $459 | 15% |
| Custom extraction | 100,000 pages a month | $2,400 | $2,040 | 15% |
| Custom extraction | 500,000 pages a month | $10,500 | $8,925 | 15% |
| Custom extraction | 1,000,000 pages a month | $18,000 | $15,300 | 15% |
| Prebuilt models | 20,000 pages a month | $190 | $152 | 20% |
| Prebuilt models | 100,000 pages a month | $900 | $720 | 20% |
| Prebuilt models | 500,000 pages a month | $4,000 | $3,200 | 20% |
| Prebuilt models | 1,000,000 pages a month | $7,500 | $6,000 | 20% |
Read the last column carefully, because it runs against the usual assumption. Containers save you the least on the cheapest model and the most on the expensive one: 10 percent on Read, 15 percent on custom extraction, 20 percent on prebuilt models. If you were considering containers to shave the cost of high-volume plain-text OCR, that is the weakest case Azure offers. If you are running prebuilt invoice or ID extraction at volume, it is the strongest.
A disconnected container runs with no path back to Azure, which is what a genuinely air-gapped environment requires. It is priced per year rather than per month, and the arithmetic is clean: every disconnected annual fee is exactly twelve times the connected monthly fee for the same tier. Nine tiers, nine exact matches. You are paying for the air gap itself, not receiving a discount for prepaying.
| Model | Tier | Connected, per month | Disconnected, per year | Sold in US regions |
|---|---|---|---|---|
| Read | 2,000,000 pages a month | $1,080 | $12,960 | Yes |
| Read | 8,000,000 pages a month | $3,780 | $45,360 | Yes |
| Read | 16,000,000 pages a month | $6,480 | $77,760 | Yes |
| Read | 500,000 pages a month | $338 | Listed in 5 regions only | No |
| Prebuilt models | 20,000 pages a month | $152 | Italy North, Poland Central, South Africa North only | No |
| Prebuilt models | 100,000 pages a month | $720 | $8,640 | Yes |
| Prebuilt models | 500,000 pages a month | $3,200 | $38,400 | Yes |
| Prebuilt models | 1,000,000 pages a month | $6,000 | $72,000 | Yes |
| Custom extraction | 20,000 pages a month | $459 | Italy North, Poland Central, South Africa North only | No |
| Custom extraction | 100,000 pages a month | $2,040 | $24,480 | Yes |
| Custom extraction | 500,000 pages a month | $8,925 | $107,100 | Yes |
| Custom extraction | 1,000,000 pages a month | $15,300 | $183,600 | Yes |
The 20,000-page disconnected tiers for prebuilt and custom models appear in exactly three regions in Microsoft's entire retail feed: Italy North, Poland Central and South Africa North. The 500,000-page disconnected Read tier appears in five. Every other disconnected tier is listed in 34 regions. No US region carries the small ones. The practical effect for a US team with an air-gap requirement is that the entry price is the 100,000-page tier, so the floor is $8,640 a year for prebuilt models and $24,480 a year for custom extraction, whether or not you will use anything like that volume. Budget from the tier you can actually buy, not the cheapest one in the price list.
Worth saying plainly, because it cuts against our own interest: if an air gap is a hard requirement for you, Azure disconnected containers are a real answer and DocuOCR is not. DocuOCR is a cloud product. We work through what does and does not qualify as on-premise across vendors on our on-premise OCR software comparison, and if the concern is document retention rather than network isolation, zero data retention OCR is usually the cheaper way to satisfy it.
A commitment tier is a fixed monthly fee that buys a page allowance, and it bills whether you use the allowance or not. So the only question that matters is how much of the tier you have to consume before it beats paying per page. The answer varies enormously by model, and it is far less forgiving on prebuilt models than on Read.
| Model | Tier | Monthly fee | Effective per 1,000 | Break-even vs pay as you go | Tier used at break-even |
|---|---|---|---|---|---|
| Read | 500,000 pages | $375 | $0.75 | 250,000 pages | 50% |
| Read | 2,000,000 pages | $1,200 | $0.60 | 800,000 pages | 40% |
| Read | 8,000,000 pages | $4,200 | $0.53 | 5,500,000 pages | 69% |
| Read | 16,000,000 pages | $7,200 | $0.45 | 10,500,000 pages | 66% |
| Prebuilt models | 20,000 pages | $190 | $9.50 | 19,000 pages | 95% |
| Prebuilt models | 100,000 pages | $900 | $9.00 | 90,000 pages | 90% |
| Prebuilt models | 500,000 pages | $4,000 | $8.00 | 400,000 pages | 80% |
| Prebuilt models | 1,000,000 pages | $7,500 | $7.50 | 750,000 pages | 75% |
| Custom extraction | 20,000 pages | $540 | $27.00 | 18,000 pages | 90% |
| Custom extraction | 100,000 pages | $2,400 | $24.00 | 80,000 pages | 80% |
| Custom extraction | 500,000 pages | $10,500 | $21.00 | 350,000 pages | 70% |
| Custom extraction | 1,000,000 pages | $18,000 | $18.00 | 600,000 pages | 60% |
We compared the published overage rate against the effective in-tier rate for every model family, every tier and both channels: 24 combinations in total. All 24 match. The 100,000-page prebuilt tier is $900 a month, which is $9.00 per 1,000 pages, and the overage rate is $9.00 per 1,000 pages. Overage on Azure is not a penalty rate, it is the same rate. That means 100 percent of the risk in a commitment tier sits on the other side: pages you paid for and did not send.
At $190 a month for 20,000 pages, the effective rate is $9.50 against a list price of $10.00. You need to use 19,000 of the 20,000 pages, every month, just to break even, and the most you can ever save is 5 percent. Unless your volume is genuinely flat, the smallest prebuilt tier is close to a rounding error with downside. The Read tiers are the opposite: the 16,000,000-page tier saves 70 percent per page at full use.
A note on the arithmetic, since it is easy to get wrong: Read drops from $1.50 to $0.60 per 1,000 pages above a million pages a month, so break-even for the larger Read tiers has to be computed in two steps. The 8,000,000-page tier at $4,200 does not break even at 2.8 million pages, it breaks even at 5.5 million, because the first million pages already cost $1,500 at list. Prebuilt and custom have no volume break, so straight division is correct for those.
No. Every batch meter matches its synchronous twin to the cent. This catches people out because batch endpoints on most LLM APIs come with a 50 percent discount, so teams assume Azure works the same way and build a queue to chase a saving that does not exist. Batch on Azure is an operational choice about how you submit work, not a pricing one.
| Model | Synchronous meter | Rate | Batch meter | Rate | Difference |
|---|---|---|---|---|---|
| Read | S0 Read Pages | $1.50 | S0 Batch Read Pages | $1.50 | Identical |
| Prebuilt models | S0 Pre-built Pages | $10.00 | S0 Batch Pre-built Pages | $10.00 | Identical |
| Layout | No synchronous Layout meter, bills as prebuilt | $10.00 | S0 Batch Layout Pages | $10.00 | Identical |
| Document classification | S0 pages for doc classifier | $3.00 | S0 Batch Document Classifier Pages | $3.00 | Identical |
| Custom extraction | S0 Custom Pages | $30.00 | S0 Batch Custom Extraction Pages | $30.00 | Identical |
| Custom generative | S0 Custom Generative Pages | $30.00 | S0 Batch Custom Generative Pages | $30.00 | Identical |
| Add-ons | S0 Add-on for Pages | $6.00 | S0 Batch Add-On Pages | $6.00 | Identical |
| Query fields | S0 pages for query fields | $10.00 | S0 Batch Pages for Query Fields | $10.00 | Identical |
There is one structural asymmetry in that table worth pointing out. Batch has a dedicated S0 Batch Layout Pages meter, but there is no synchronous Layout meter at all: call Layout synchronously and it bills on S0 Pre-built Pages, alongside invoice and receipt extraction. The batch Layout meter took effect on 2026-01-01, which makes it the newest meter in the product. Same $10.00 either way, but if you are reconciling an invoice line by line, that is why Layout appears under two different names depending on how you called it.
Pay-as-you-go, no commitment tier, US East. The Read column includes the $0.60 volume break above one million pages a month, which is why the five million page row is $3,900 rather than $7,500.
| Volume | Read (plain text) | Layout or a prebuilt model | Custom extraction |
|---|---|---|---|
| 10,000 pages a month | $15 | $100 | $300 |
| 50,000 pages a month | $75 | $500 | $1,500 |
| 250,000 pages a month | $375 | $2,500 | $7,500 |
| 1,000,000 pages a month | $1,500 | $10,000 | $30,000 |
| 5,000,000 pages a month | $3,900 | $50,000 | $150,000 |
These are API costs and nothing else. They do not include the classification step that routes a mixed inbox to the right model, the validation rules that catch a misread total, the human review queue for anything below your confidence threshold, storage, or the engineering time to build and keep all of it running. That gap between an API rate and a working document workflow is the honest subject of our OCR API pricing comparison, and if you want the same rates lined up against AWS, Google, Mistral and the LLM readers, OCR pricing per 1,000 pages is the cross-vendor reference.
The per-meter rate card in full, including which model ID bills on which meter and why an invoice can still say Form Recognizer.
Read itAzure OCR is two different products on two different meters, and Microsoft recommends against one of them. Which to call, and what each costs.
Read itFile size, page ceilings and rate limits, which decide your architecture long before price does.
Read itThe same treatment applied to Amazon: per-meter rates, the multi-feature discount, and the 1.40x region trap.
Read itMicrosoft's newer multimodal service, on its own meters, priced against Document Intelligence.
Read itWhat you give up and what you gain by moving off Azure, including the work the API rate does not cover.
Read itA rate card tells you what a page costs. It cannot tell you how many pages you will have to send twice, or how many fields a human will have to fix. Upload the document you were about to test on Azure and see the finished, validated output.
