Azure Document Intelligence File Size Limit: How Big Can a Document Be?

Last updated July 2026.

The Azure Document Intelligence file size limit is 500 MB on the paid Standard (S0) tier and 4 MB on the free F0 tier. A single analyze call also caps at 2,000 pages on S0. Both the size and page limits are set limits that cannot be raised through a support request, so a file over 500 MB has to be split before you send it. The free F0 tier is stricter than the 4 MB number suggests: it returns results for only the first two pages of any request.

This is one of the first numbers you need when you plan an Azure AI Document Intelligence integration, and it is easy to get wrong because the limit depends on your pricing tier and the free tier hides a second surprise. The figures below are read from Microsoft's Document Intelligence service-limits documentation in July 2026. Microsoft updates that page, so confirm the current values before you design around them.

What is the maximum file size for Azure Document Intelligence?

The maximum document size is 500 MB on the paid Standard (S0) tier and 4 MB on the free F0 tier. That 500 MB ceiling is the largest single-file size in the cloud OCR group, level with AWS Textract's asynchronous limit and far above Google Document AI's 40 MB online cap. Alongside the size limit, one analyze request accepts up to 2,000 pages on S0. Neither the size nor the page limit is adjustable, so they set a hard boundary on any single call, no matter how much you are willing to pay.

The practical read is simple. On S0 you can hand Azure a 400-page scanned PDF that weighs a few hundred megabytes and get it back in one call, with no asynchronous job and no storage bucket in the middle. That single-call ceiling is the main reason teams pick Azure over Textract for long documents, since Textract's synchronous call reads one page and forces an async workflow for anything more.

Why is my Azure Document Intelligence file too large?

If Azure rejects a file as too large, it is over the tier limit: 4 MB on F0 or 500 MB on S0. High-resolution color scans are the usual cause on the free tier, where a handful of 300 DPI pages can blow past 4 MB quickly. On S0 the 500 MB limit is generous enough that only very long or very high-resolution documents reach it. When you do, the answer is to split the file, because the limit cannot be raised.

There is a second, quieter failure that is not a size error at all. On F0 the analyze response returns only the first two pages of any document. Pages three and beyond come back empty, which looks like a model failure rather than a tier limit. If your file is under 4 MB and Azure still seems to stop reading after page two, you are on F0 and need to move to S0.

How does Azure count pages against the limit?

How Azure counts a page depends on the file type, and it affects both your 2,000-page ceiling and your bill. For PDF and TIFF files, each page or image counts as one page, with no character cap. For Microsoft Word and HTML files that the Read and Layout models support, Azure counts pages in blocks of 3,000 characters, so a 7,000-character document counts as three pages. For Excel and PowerPoint, each worksheet and each slide counts as one page.

Format support has its own catch. All models read PDF, JPEG, PNG, BMP, TIFF and HEIF. Microsoft Office formats, DOCX, PPTX and XLS, are only supported by the Read and Layout models, not by the prebuilt or custom models. So you can OCR a Word file, but to run it through the prebuilt invoice model or a custom extraction model you convert it to PDF first. That conversion step is easy to miss until a prebuilt call fails on an Office file.

File size limit by tier

For the full breakdown of every quota, including the rate limits and custom-model training caps, see the Azure Document Intelligence limits reference.

How to process a document over the size limit

If a file is over 500 MB, you split it. Break the PDF into chunks under the limit, process each chunk, and reassemble the results by page number. The same approach handles documents over 2,000 pages: process in 2,000-page batches and stitch the output. The one thing to watch is tables and fields that span a chunk boundary, since a split in the middle of a multi-page table can break the extraction and needs a rule to merge the halves.

Not every workload needs the raw document structure, though. If your real goal is to get the numbers out of a long statement and into a workbook, a dedicated PDF to Excel converter gets you a clean spreadsheet without touching an OCR API at all. When you do need field-level extraction across many long files, the splitting, batching and reassembly is the part a ready-to-use product handles for you: you upload the whole document, it manages the size and page limits, and it returns the fields.

Is the Azure Form Recognizer file size limit different?

No. Azure Form Recognizer is the former name of Azure AI Document Intelligence. Microsoft renamed the service but kept the same limits and the same REST API, so the maximum file size is still 500 MB on S0 and 4 MB on F0. Any older guide or Stack Overflow answer that refers to Form Recognizer is describing the current Document Intelligence limits. If you see both names in your own code and docs, they point at the same service.

How Azure compares with other OCR APIs on file size

File size is one place the major OCR services diverge sharply. Azure and Textract sit at the top with 500 MB ceilings, while Google's online endpoint caps at 40 MB and the large-language-model readers cap lower still. The table below shows the maximum file size and single-request page count for each, verified in July 2026.

The pattern worth noting: Azure gives you the highest single-call page count with no asynchronous plumbing, while Google forces a batch job past 15 pages and the LLM readers share a lower 50 MB ceiling that bites on large scans. See how the ceilings stack up in the OCR API limits comparison, and the per-1,000-page rates in the Azure Document Intelligence pricing breakdown.

The short version

Azure Document Intelligence reads files up to 500 MB and 2,000 pages in one call on S0, and 4 MB on the free F0 tier, which also returns only the first two pages of a request. The limits cannot be raised, so oversized files get split, and real work belongs on S0 rather than the F0 sandbox. If you would rather not manage tiers, splitting and page ranges at all, DocuOCR reads a long document in one upload and returns clean fields. Drop in your longest file and see it come back whole.