Azure Content Understanding adds synchronous processing preview for document extraction

Microsoft has added synchronous processing options to Azure Content Understanding, the document-extraction service that ships as part of Foundry Tools. The Read, Layout, and Digital Parse analyzers now support a public-preview mode that extracts content from a document and returns structured results directly in the HTTP response, without staging the input in temporary service-side storage.

The change addresses two needs. Agentic workflows — grounding AI agents on uploaded documents, verifying identities, helping call center agents handle customer paperwork — often require extracted content before a response can be generated. The asynchronous path requires submitting a job and polling for results, which adds latency. Synchronous requests instead use priority processing and resources designed for lower-latency scenarios. Separately, Microsoft says enterprise customers using both Azure Document Intelligence and Azure Content Understanding asked for an option that avoids temporary service-side storage when handling highly confidential documents; the new operations process input in memory instead.

The Read analyzer uses OCR to extract text and location data from images and PDFs. Layout goes further, recovering document structure such as tables, sections, figures, format information, hyperlinks, and now signatures from more than 30 file types. Synchronous requests accept a document either as binary data or via an accessible URL, and support documents up to 10 MB, five pages, or 30,000 characters per request, with an option to specify a page range. File formats match those of asynchronous operations.

Pricing sits at a premium over the asynchronous tier: Read costs $1.50 per 1,000 pages (versus $1.00 asynchronously), Layout $7.50 per 1,000 pages (versus $5.00), and text documents $0.015 per 1,000 pages (versus $0.01), with up to 3,000 characters counted as one page.

The preview has clear boundaries. Only the three prebuilt analyzers — prebuilt-read, prebuilt-layout, and prebuilt-digitalParse — support synchronous calls; all other prebuilt analyzers remain asynchronous-only. Microsoft notes that all preview features, limits, pricing, API versions, and SDK details may change before general availability.

Developers can invoke the operations through REST endpoints (analyzeInline for URL inputs, analyzeBinaryInline for binary payloads) or through updated SDKs for Python, .NET, JavaScript, and Java. The Python SDK requires an --pre install flag to pull in preview features; Microsoft's samples show a synchronous Read call on an insurance card image and a Layout call on an invoice PDF.

The synchronous path complements the asynchronous Content Understanding operation, which has been generally available since November 2025. For teams processing sensitive documents, the preview offers an alternative that trades the five-page, 10 MB request limits and higher per-page pricing for immediate results and no server-side storage.