Beyond the News
Sign in
Artificial IntelligenceCloud & InfrastructureCloud PlatformsInference & APIs Active dossier

Azure Document Intelligence v2.0 API retires August 31 — old integrations need a version migration

Microsoft will retire the Azure Document Intelligence v2.0 API on August 31, 2026. Applications still calling that REST version need to move to a supported API before the cutoff to avoid service disruption.

Reviewed and published by Lars Jansen Updated 5 min read
First published

Key details

  1. Azure Document Intelligence v2.0 is scheduled to retire on August 31, 2026.
  2. Microsoft recommends Azure Document Intelligence v4.0, API version 2024-11-30, for current development.
  3. Microsoft redesigned the Document Intelligence REST API starting with v3.0, so v2.x migrations can require code changes beyond changing a version string.
  4. The safest impact check is to identify the actual REST api-version used by each production integration.
  5. Later API versions have different retirement dates; the August 31, 2026 deadline specifically targets v2.0.

What builders should take away

  1. Search code and configuration for Azure Form Recognizer or Document Intelligence calls that explicitly set an old api-version, including direct REST clients and wrappers that may not be obvious from package names.
  2. Capture a production request or enable HTTP/SDK logging to verify the exact api-version currently sent before deciding an application is unaffected.
  3. Migrate one representative workflow to v4.0 and replay real document samples, comparing extracted fields, confidence values, page/layout structure, latency and error behavior.
  4. Test custom models and downstream parsers explicitly; a successful HTTP response is not enough if the result schema or field semantics changed.
  5. Move production traffic before August 31 and monitor error rates plus extraction-quality regressions during the cutover.

What changed

Microsoft’s long-announced retirement date for Azure Document Intelligence v2.0 is now less than two weeks away. The v2.0 API is scheduled to retire on August 31, 2026, and Microsoft’s current migration guidance recommends the generally available v4.0 API for new development and for workloads moving off older versions.

Why it matters

Document Intelligence sits inside production document-processing pipelines for invoices, receipts, forms and custom extraction. After a retired API version stops accepting requests, the failure is operational rather than cosmetic. The migration also crosses a redesigned REST surface introduced after v2.x, so teams should test request formats, model behavior, result schemas and SDK-generated api-version values rather than treating this as a simple endpoint rename.

The v2.0 cutoff is August 31

Microsoft’s Azure retirement notice states that Document Intelligence v2.0 will retire on August 31, 2026. Teams that still send requests against that version should treat the remaining window as a production migration deadline, especially where document extraction sits in billing, onboarding, compliance or back-office workflows.

Microsoft points builders to v4.0

Microsoft’s migration guide describes Document Intelligence v4.0, using API version 2024-11-30, as the latest generally available version and recommends it for the richest feature, language and document-type coverage. The guide explicitly covers migration from v2.0 and v2.1 and notes that the REST API was redesigned starting with v3.0.

Check the wire-level api-version, not just the SDK name

The practical risk is hidden dependency on an old REST version. An application can appear to use a current-looking client package while still being configured to call an older endpoint or api-version. Inventory direct REST calls, generated clients, wrappers and long-lived services, and inspect actual request URLs or SDK configuration to establish which API version production traffic uses.

Expect behavior-level migration work

Because the API shape changed after v2.x, teams should compare request payloads, operation names, asynchronous polling, result fields, custom-model references and error handling under the target version. Representative production documents should be replayed through the new path before switching traffic.

Timeline

2023-09-15

Microsoft announces v2.0 retirement

Microsoft publishes the retirement notice setting August 31, 2026 as the end date for Azure Document Intelligence v2.0.
2026-08-31

v2.0 retirement date

Azure Document Intelligence v2.0 is scheduled to retire.

What to watch next

  • Whether Microsoft changes the August 31 retirement date or publishes final migration advisories before the cutoff.
  • Reports of compatibility issues involving custom models, SDK defaults or result-schema differences during v2.x-to-v4.0 migrations.
  • Any additional near-term Azure AI API retirements that affect the same document-processing stacks.

Still unclear

  • Microsoft’s public documentation does not provide a universal mapping from every historical SDK package version to the REST api-version it ultimately sends, so teams may need to inspect live configuration or request logs to prove impact.
  • The amount of migration work varies substantially by workload; simple prebuilt-model calls may be easier than custom-model pipelines with downstream schema assumptions.

Sources

Direct reading behind this dossier.

3 sources
Continue reading
Artificial Intelligence

OpenAI lets one API project choose regional processing per request

OpenAI’s August 21 control moves processing-region choice into request routing: a single Global project can send eligible calls to regional base URLs. That simplifies multi-region SaaS architecture, but builders still need to enforce residency policy in code and account for support, retention and pricing constraints.

Artificial Intelligence

Cloud Run sandboxes now work inside jobs and worker pools

Cloud Run sandboxes now cover all resource types. The August 5 expansion matters for builders whose agents or automation run in batch jobs or continuously pulling workers rather than HTTP services, while the feature remains pre-GA and shares CPU and memory with the host container.