Azure Document Intelligence reads up to 500 MB and 2,000 pages in one call on the paid S0 tier, with a 15 analyze-per-second rate limit. The free F0 tier caps files at 4 MB and returns only the first 2 pages.
The free-tier two-page cap is the surprise that makes evaluators think the model failed. Here is every Azure limit, read from Microsoft's docs. Last updated July 2026.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Uploading...
Read a long document without picking a tier or a page range.
Azure AI Document Intelligence, the service Microsoft used to call Form Recognizer, has two sets of limits that depend entirely on your pricing tier. On the paid Standard (S0) tier a single analyze call accepts a document up to 500 MB and 2,000 pages, and neither figure can be raised. That is the highest single-request page count in the cloud OCR group, so a 500-page PDF goes through in one call with no asynchronous job. The rate limit is 15 analyze transactions per second by default, plus 50 get-operations per second for polling results, and both are adjustable for free through an Azure support request. The free F0 tier is where teams get caught: it caps files at 4 MB, holds every per-second limit at 1, and returns results for only the first 2 pages of any request. Everything here is read from the Microsoft documentation in July 2026, and because Microsoft updates its quotas, confirm each figure on the current service-limits page before you build around it.
The most common Azure Document Intelligence support question is not about the 500 MB ceiling. It is a developer testing on the free F0 tier who sends a ten-page invoice and gets fields back for pages one and two, then nothing. The model looks broken. It is not. F0 returns analyze results for only the first two pages of any request, by design, so you can test features without processing real documents at no cost.
The fix is to move to the Standard S0 tier, where the same call reads up to 2,000 pages. There is no code change and no different endpoint, only the resource tier. Because F0 and S0 share the REST API, a proof of concept built on F0 will suddenly read the whole document the moment you point it at an S0 resource. If pages three and beyond come back empty, check your tier first.
The second free-tier limit is the 4 MB file cap versus 500 MB on S0, and every per-second transaction limit fixed at 1. F0 is a feature sandbox, not a low-volume production tier. If you need to process even a handful of real multi-page files, you are on S0 from the start.
Max document size and max pages are set limits and cannot be raised. Analyze and get TPS can be increased free via a support ticket. Verified from Microsoft docs, July 2026.
The same limits split by tier, because which tier you are on decides your page ceiling, file size and rate.
| Limit | Free (F0) | Standard (S0) |
|---|---|---|
| Max document size | 4 MB | 500 MB |
| Max pages per analyze call | First 2 pages returned | 2,000 pages |
| Analyze transactions / second | 1 | 15 (default, adjustable) |
| Get operations / second | 1 | 50 (default, adjustable) |
| Model-management ops / second | 1 | 5 (default) |
| List operations / second | 1 | 10 (default) |
| Max OCR JSON response | 500 MB | 500 MB |
See how these ceilings compare with AWS and Google on the OCR API limits comparison, and what Azure costs per 1,000 pages on the Azure Document Intelligence pricing breakdown.
These rules apply on every tier and decide what Azure will read and how it counts a page toward your ceiling.
| Constraint | Rule |
|---|---|
| Accepted formats (all models) | PDF, JPEG/JPG, PNG, BMP, TIFF, HEIF. |
| Office formats (Read and Layout only) | DOCX, PPTX, XLS. Not supported by prebuilt or custom models. |
| PDF and TIFF page counting | Each page or image counts as one page, no character cap. |
| Word and HTML page counting | Counted in blocks of 3,000 characters per page. |
| Excel and PowerPoint | Each worksheet or slide counts as one page. |
| Custom classifier training | Up to 25,000 pages and 1,000 document classes. |
| Custom neural training | 10 free hours per month, then billed by the hour. |
The Office-format gap catches teams too: you can OCR a DOCX or XLS with Read and Layout, but a prebuilt or custom model needs a PDF. An Azure Document Intelligence alternative that takes mixed file types and long PDFs in one upload removes the tier and format math.
Azure next to Textract, Google, Mistral and Gemini.
The 1-page sync vs 3,000-page async split.
Every Azure model rate per 1,000 pages.
Skip the tier math and Office-format gap.
How the two clouds compare on limits and output.
Single-call 2,000 pages vs async 3,000.
Per-1,000-page rates across the market.
An honest roundup for US teams.
Google's 15-page online cap and the batch fix.
Azure's free tier reads two pages and its Office support skips the prebuilt models. DocuOCR takes a mixed batch or a long PDF in one upload, reads every page, and returns clean fields. Drop in your longest document and compare.
