Azure/azure-sdk-for-net Azure.AI.ContentUnderstanding_1.2.0-beta.3

on GitHub

1.2.0-beta.3 (2026-08-11)

Features Added

• Added support for service API version 2026-06-01-preview, which is the default service API version for this beta package.
• Added inline analysis convenience APIs AnalyzeInline / AnalyzeInlineAsync and AnalyzeBinaryInline / AnalyzeBinaryInlineAsync, available only with service API version 2026-06-01-preview. These return AnalysisResult in a single HTTP 200 response (no LRO polling) and throw RequestFailedException when the inline operation status is not Succeeded. AnalyzeBinaryInline* includes a ContentRange? convenience overload matching AnalyzeBinary*. See Sample 18 and Sample 19.
• Added AnalyzeBinaryOptions and corresponding AnalyzeBinary* / AnalyzeBinaryInline* overloads for binary analyze request settings (ContentRange, ContentType, ProcessingLocation, and future options). Required analyzer ID and binary input live on the options bag. See Sample 01 and Sample 19.
• Added AnalyzeOptions and corresponding Analyze* / AnalyzeInline* overloads for JSON analyze request settings (ModelDeployments, ProcessingLocation, and future options). Required analyzer ID and inputs live on the options bag.
• Added semantic chunking for custom analyzers in 2026-06-01-preview: configure ContentAnalyzerConfig.ChunkingStrategy with SemanticChunkingStrategy (for example MaxTokens) when creating an analyzer, then read DocumentContent.Chunks (DocumentChunk spans into markdown) from the analysis result. See Sample 17.
• Added analyzer workflow selection via ContentAnalyzerConfig.Workflow / ContentAnalyzerWorkflow for 2026-06-01-preview. Omit Workflow for standard extraction, or set ContentAnalyzerWorkflow.Agentic when an answer must be built from evidence. See Sample 16.
• Added signature detection via DocumentSignature / DocumentContent.Signatures for 2026-06-01-preview when layout extraction is enabled (EnableLayout, including prebuilt-layout). See Detect signatures and Sample 10.
• Added in-page segmentation opt-in via ContentAnalyzerConfig.AllowInPageSegments for 2026-06-01-preview. Used with EnableSegment, this allows classification segments to split within a page (for example supplemental statements appended after a K-1 tax form) instead of only at page boundaries. See Classify in-page segments.
• Added embedded document metadata via AnalysisContent.Metadata for 2026-06-01-preview. See Extract document metadata.
• Added analysis diagnostics via AnalysisResult.Infos for 2026-06-01-preview. The collection exposes service information as ResponseError values for troubleshooting. See Analysis diagnostics.
• Updated ToLlmInput (preview) to emit analysis-result metadata (AnalysisContent.Metadata) under a metadata: front-matter block. See ToLlmInput.
• Added AnalyzeOperationExtensions.GetUsageDetails() to return generated UsageDetails from a completed analyze LRO (Operation<AnalysisResult>) or inline analyze response (Response<AnalysisResult>). See Sample 03, Sample 18, and Sample 19. GetUsage() / AnalyzeUsageDetails are obsolete and retained for 1.1.0 compatibility.

Other Changes

• ToLlmInput (preview): renamed the optional caller dictionary from metadata to customMetadata and emit it under a nested customMetadata: front-matter block (service metadata stays under metadata:). Changes the preview API from 1.2.0-beta.1; not a stable-breaking change.