Microsoft now sells two document AI services, and the newer one is cheaper on every extraction meter. That does not make it the right answer for everyone. Here is what each one actually does, where each wins, and whether a migration is worth your time.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Uploading...
Neither Azure service ships a review screen or an export. Drop a document in here and see what a finished pipeline returns instead.
Use Azure Content Understanding if you are paying for OCR or layout at volume, if a meaningful share of your files are digital Word, Excel or email documents, or if you also need to process audio and video. It is 33% cheaper for OCR, 50% cheaper for layout, and 150 times cheaper on digital files, because it bills those on a Minimal meter at $0.01 per 1,000 pages instead of running OCR they never needed. Stay on Azure AI Document Intelligence if you need a flat and forecastable per-page rate, if you send documents longer than 300 pages, if you route high volumes through the $3 per 1,000 pages classifier, or if the prebuilt invoice and W-2 models already return what you need. Microsoft has not deprecated Document Intelligence, has announced no end of support, and its meters are still live, so there is no deadline forcing this decision.
Document Intelligence is a self-contained service: one call, one per-page rate, one bill.
Content Understanding is an orchestration layer that runs generative work on a Foundry model deployment you provision. Those tokens are billed to that deployment, not to the service.
Cheaper at the meter. Two bills. And a cost that moves with the model you pick.
No. Microsoft has announced no end of support for Azure AI Document Intelligence, and every one of its billing meters is still live and active in the official Azure price feed, including meters that took effect as recently as January 2026. The service is being sold and supported.
What is genuinely true, and worth acting on, is narrower: two preview API versions of Content Understanding are being retired. That is a Content Understanding deadline, not a Document Intelligence one, and the two keep getting conflated in blog posts. Nothing on the Document Intelligence side has a published sunset.
Content Understanding API versions 2024-12-01-preview and 2025-05-01-preview are being retired on July 15, 2026.
The generally available version is 2025-11-01. If your code still points at a preview endpoint, move it now. The GA release also changed how generative features are billed, so re-check your cost model while you are in there rather than assuming the old estimate still holds.
Document Intelligence has no equivalent deadline. If you are on it and it works, you are not on a clock.
Rates from Microsoft's Azure Retail Prices API, limits and capabilities from Microsoft Learn, both read on July 13, 2026. The highlight marks which service wins that row, where one clearly does.
| Capability | Content Understanding | Document Intelligence |
|---|---|---|
| Documents (PDF, TIFF, images) | Yes | Yes |
| Audio (speech to text) | Yes, $0.36 an hour | No |
| Video (frames, shots, transcript) | Yes, $1.00 an hour | No |
| Images and text files | Yes | Documents only |
| Plain OCR rate | $1.00 per 1,000 pages | $1.50 per 1,000 pages |
| Layout, tables, structure | $5.00 per 1,000 pages | $10.00 per 1,000 pages |
| Digital DOCX, XLSX, HTML, TXT | $0.01 per 1,000 pages | $1.50 (no digital meter) |
| Custom field extraction | Schema in JSON, your model, variable cost | $30.00 per 1,000 pages, flat |
| Document classifier | Contextualization plus model tokens | $3.00 per 1,000 pages, flat |
| Confidence scores and source grounding | First-class output | Available on trained models |
| Improve accuracy without retraining | Yes, a few labeled examples in context | Requires training a model |
| Max pages in one call | 300 | 2,000 |
| Max file size | 200 MB | 500 MB |
| Needs a Foundry model deployment | Yes, for anything generative | No |
| Number of bills | Two: the service and your model | One |
| Batch discount | None | None |
All rates per 1,000 pages unless stated. The full rate card for the newer service, including contextualization tokens and Microsoft's own worked cost examples, is on our Azure Content Understanding pricing reference.
This is the clearest win. The same OCR is $1.00 per 1,000 pages instead of $1.50, and layout is $5.00 instead of $10.00. On 5 million pages of layout a year, that is $25,000 rather than $50,000, for identical work.
Word, Excel, PowerPoint, HTML, text and email files bill on the Minimal meter at $0.01 per 1,000 pages, because no OCR is performed. Document Intelligence has no digital meter and charges the full Read rate of $1.50 to read a file that was never a scan. That is a 150-fold difference.
Content Understanding lets you define a schema in JSON and improve it with a handful of labeled examples in context, without retraining anything. On documents that vary by supplier or by state, that iteration loop is far shorter than training and redeploying a custom model.
Confidence scores and source grounding, tracing every field back to where it appeared on the page, are first-class output. If a human reviews low-confidence fields, that is the signal that decides what reaches them.
Audio and video content extraction is not something Document Intelligence can do at any price. If call recordings or video sit in the same pipeline as your documents, one service now covers all of it.
A cheaper meter is not automatically a cheaper project. Four cases where the older service is genuinely better, not just familiar.
Document Intelligence publishes a rate for every model and bills nothing else. Content Understanding's generative cost moves with your schema, your feature flags and the model you attach, which is harder to put in a budget.
Content Understanding caps a single document at 300 pages and 200 MB. Document Intelligence takes 2,000 pages and 500 MB, the highest single-call ceiling of any major vendor.
The Document Intelligence classifier is $3.00 per 1,000 pages, flat. On Content Understanding, categorization is a generative feature, so it costs contextualization plus model tokens on every page you route.
If your invoices, receipts, W-2s and 1099s already come back correctly from a Document Intelligence prebuilt model at $10.00 per 1,000 pages, a migration buys you very little and adds a model deployment to run.
Every meter, official rates, and Microsoft's own worked cost examples.
Read itFull rate card plus the commitment tiers Microsoft does not publish.
Read itEvery vendor normalized onto one unit, including per-document and per-token pricing.
Read itThe other Azure comparison buyers run before they commit.
Read itPage ceilings, file size caps and the F0 free-tier behavior that trips up evaluations.
Read itWhat to use if neither Azure service fits how your team works.
Read itWhichever Azure service you land on, the output is an API response. Routing each document to the right analyzer, reviewing the fields the model was unsure about, applying your validation rules and pushing the result into your accounting system or ERP are all still yours to build and maintain. On most projects that work costs more than the meter ever will.
DocuOCR charges more per page, roughly $14 to $20 per 1,000, and ships that pipeline finished. If your team would rather own the integration, an Azure API is the cheaper path. If not, this is what the alternative looks like.
Upload a document you actually process and look at the fields that come back. It is the only comparison that predicts what a migration will cost you in review time.
Extract a document freeLast updated July 2026. Verified against Microsoft Learn and the Azure Retail Prices API.
Z tej samej rodziny narzędzi
