Implement information extraction solutions 5.2 Extract content from documents

Your team used Content Understanding pro mode to validate invoice totals across line…

Study guide skill: Implement analyzers for generating structured or markdown outputs for downstream reasoning by using Content Understanding

Question B2-030

Your team used Content Understanding pro mode to validate invoice totals across line items. The pro-mode API version has been retired. What should you use now for this reasoning-heavy extraction?

Select one answer.

  • A A video analyzer with custom segments
  • B The prebuilt-read analyzer
  • C Keep calling the 2025-05-01-preview API with pro mode
  • D A document analyzer created with the 2026-06-01-preview API and config.workflow set to agentic (agentic mode, preview) ✓ Correct answer

Answer: D

Pro mode was available only in the 2025-05-01-preview API, which is retired, and the documented migration path is agentic mode (preview) in the 2026-06-01-preview API, enabled by setting config.workflow to agentic on a document analyzer. Agentic mode is designed for multistep reasoning, calculations, and validation, but it isn't a drop-in replacement for pro mode, so test it on representative documents. prebuilt-read only extracts text, video analyzers don't analyze invoices, and the retired API can't be used.

Deep dive

How it works

Agentic mode (preview) is Content Understanding's option for documents where the answer must be built from evidence rather than extracted from one spot. It can reason across a document, perform calculations, validate results and return structured fields that match your schema. Validating invoice totals across line items is a textbook case: calculations (totals and reconciliations) plus validation (internal consistency).

The migration facts on the agentic mode page:

  1. Pro mode is only available in the 2025-05-01-preview API, which is now retired.
  2. Migrate to agentic mode in the 2026-06-01-preview API, by setting config.workflow to "agentic" when you create a document analyzer. After creation, the service returns a resolved, versioned value.
  3. It isn't a drop-in replacement for the pro mode configuration. Review your analyzer and field schema first.

Plan for the preview limits and cost. Each request supports one input file, fields that use the extract method aren't supported, and agentic mode typically needs about 400,000 TPM per analyzer job on the model deployment. Test with representative documents to confirm the quality gain is worth the latency and cost.

The other options, and when you'd use them

  • A video analyzer with custom segments: Agentic mode supports document analyzers only, and an invoice is a document. Video analyzers are for video content.
  • The prebuilt-read analyzer: prebuilt-read extracts words, paragraphs, formulas and barcodes with basic OCR. It provides text extraction without layout analysis, and no reasoning or calculations.
  • Keep calling the 2025-05-01-preview API: That API is retired, so there's nothing to keep calling. The documented path is agentic mode in the newer preview API.

Exam tip

Content Understanding "reason, calculate, validate" = document analyzer with config.workflow "agentic" (2026-06-01-preview). Pro mode lived only in the 2025-05-01-preview API.

References

Microsoft Learn references

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Information extraction questions · AI-103 study guide

Thanks for your feedback!