Understanding Major AI and Cognitive API Retirements

Question

Understanding Major AI and Cognitive API Retirements

Luke Feetterer 0

Hello,

I've been seeing that there are several API retirements upcoming very soon and I'm having a hard time fully understanding if my team is impacted, as we're using mostly SDK's / wrappers which may or may not call these soon-to-be retired APIs downstream. Any guidance or information on how to better understand / track these retirements would be very helpful.

Azure OpenAI Assistants API (8/26/26 retirement)

We're not using this API directly, but I've been reading that any code which follows a create thread / create run etc pattern would still be affected by this. More specifically, my team is using Azure.AI.Projects and Azure.AI.Agents.Persistent, which does involve the thread / vector store / run creation pattern. Will this code stop working on 8/26/26?

Can you provide a mapping of Azure SDKs to the underlying REST APIs they depend on, and identify which SDKs (if any) require customer action because of upcoming API retirements?

Are there any other SDKs at risk? AzureOpenAIClient Chat Completions, Azure.AI.Extensions.OpenAI, etc.?

Azure AI Document Intelligence v2.0 API (8/31/26 retirement)

In a couple of products that I work on, we're using the Azure.AI.DocumentIntelligence nuget package, which only has a stable version 1.0 available. Will this be impacted?

Any information or guidance that you can provide would be greatly appreciated, it's been difficult to track some of these things down.

SRILAKSHMI C 19,730 Reputation points Microsoft External Staff Moderator

2026-09-03T15:36:34.85+00:00

Hello @Luke Feetterer

Following up to see if the below answer was helpful. If this answers your query, please do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Thank you!

2 answers

Your answer

SRILAKSHMI C 19,730 Reputation points Microsoft External Staff Moderator

2026-09-03T15:36:34.85+00:00

Hello @Luke Feetterer

Following up to see if the below answer was helpful. If this answers your query, please do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Thank you!

Answer 1

Hello @Luke Feetterer

Thank you for reaching out to Microsoft Q&A.

We understand the difficulty of tracking API retirements when your applications use Azure SDKs or wrappers rather than calling the REST APIs directly.

The key point is that an API retirement generally affects an application when the application or the SDK it uses ultimately calls the retired API operation/version. The SDK package version and the underlying service API version are separate concepts, so an SDK should not be considered impacted solely based on its package name or the functionality exposed by the SDK.

Azure OpenAI Assistants API

For the Azure OpenAI Assistants API retirement, Microsoft recommends migrating affected workloads to the Responses API.

For Azure.AI.Projects and Azure.AI.Agents.Persistent, the fact that your application uses concepts such as threads, runs, or vector stores does not by itself confirm that the application is calling the retiring Assistants API.

The documentation provided does not contain a definitive mapping between these SDKs and the specific REST API operations/versions they use. Therefore, we would not recommend concluding that these SDKs will stop working on the retirement date solely because they expose thread/run functionality.

For your specific application, we recommend:

Identify the exact versions of Azure.AI.Projects and Azure.AI.Agents.Persistent being used.

Review the corresponding SDK release notes and documentation for API-version or retirement guidance.

Where possible, inspect the HTTP traffic generated by the SDK and identify the underlying REST endpoint and API version.

If the SDK is ultimately calling a retired Assistants API endpoint/version, migrate to the supported replacement pattern before the retirement date.

The same principle applies to AzureOpenAIClient Chat Completions and Azure.AI.Extensions.OpenAI. Being an Azure OpenAI SDK does not by itself mean that the SDK is affected by the Assistants API retirement; the specific API surface and version being called need to be verified.

Document Intelligence v2.0 API

For the Azure AI Document Intelligence v2.0 API retirement, the NuGet package version should not be used as the sole indicator of impact.

For example, the fact that Azure.AI.DocumentIntelligence currently has a stable 1.0 package version does not, by itself, establish whether your application is calling the retiring v2.0 service API.

We recommend checking the actual Document Intelligence API version used by your client/application. If the SDK is sending requests to the retiring API version, an SDK update or API-version change will be required before the retirement date.

How to track Azure API and model retirements

For Azure OpenAI model retirements, Microsoft provides lifecycle notifications for subscriptions with affected deployments. Subscription owners should receive retirement notifications, and you can also review your existing deployments to determine which model and model version they use.

You can list Azure OpenAI deployments using Azure CLI:

az cognitiveservices account deployment list --resource-group <rg> --name <account>

You can also use Azure PowerShell:

Get-AzCognitiveServicesAccountDeployment -ResourceGroupName <rg> -AccountName <account>

If a deployment is using a model/version that is scheduled for retirement, it should be migrated to a supported model/version before the retirement date. Also review the deployment's upgrade policy, since the behavior can differ depending on whether the deployment is configured for automatic upgrades or is pinned to a specific model version.

Recommended approach for SDK-based applications

For applications that use SDKs or wrappers, the most reliable process is:

Inventory the Azure SDK/package names and exact versions used by your applications.

Review the SDK release notes for retirement and API-version changes.

Identify the underlying REST API version used by each SDK where possible.

Compare those API versions with Microsoft's retirement announcements.

Upgrade the SDK or migrate the API usage where the underlying API is being retired.

Test the updated application before the retirement date.

The important takeaway is that you do not need to assume that every SDK related to a retiring service will stop working. The determining factor is whether the SDK/application ultimately depends on the specific API operation or version being retired.

For the two specific SDKs you mentioned, the documentation provided does not contain enough information to give a definitive statement that Azure.AI.Projects / Azure.AI.Agents.Persistent will stop working on August 26, 2026, or that Azure.AI.DocumentIntelligence version 1.0 will be affected by the August 31, 2026 Document Intelligence API retirement.

Checking the exact SDK versions and the underlying API requests is therefore the appropriate way to establish whether customer action is required.

Please refer this

Azure OpenAI in Azure AI Foundry Models model deprecations and retirements: https://learn.microsoft.com/azure/ai-foundry/openai/concepts/model-retirements
Retired Foundry Models (Azure OpenAI): https://learn.microsoft.com/azure/foundry/openai/concepts/retired-models
Resolve deprecation of OpenAI Assistants API (Responses API / migration referenced in provided doc): https://learn.microsoft.com/azure/ai-foundry/openai/how-to/responses

I Hope this helps. Do let me know if you have any further queries.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thank you!

Answer 2

Allan Solomon Mejia 7,995

Hello @Luke Feetterer

The easiest way to determine whether you're affected is to identify the underlying REST API version that your SDK is calling, rather than relying solely on the SDK package name.

For your scenarios:

Azure OpenAI Assistants API

If your application uses the assistant → thread → run pattern (including Azure.AI.Projects or Azure.AI.Agents.Persistent against the classic Assistants architecture), you should review your implementation and plan a migration to the Microsoft Foundry Agent Service / Responses API. Microsoft's migration guidance recommends moving away from the Assistants API before its retirement.

If you're only using chat completions through AzureOpenAIClient (without Assistants, Threads, Runs, or Vector Stores), those workloads are not affected by the Assistants API retirement. The retirement is specific to the Assistants API surface, not the Chat Completions API.

Azure AI Document Intelligence

The Azure.AI.DocumentIntelligence SDK version alone doesn't determine whether you're impacted. What matters is whether your application is calling the Document Intelligence v2.0 REST API. If it is, you'll need to migrate to a supported API version before the retirement date. Review the API version configured by your client or inspect the request URI/API version in your application.

Unfortunately, Microsoft doesn't currently publish a comprehensive matrix mapping every Azure SDK to its underlying REST API versions. For applications with multiple dependencies, the most reliable approach is to:

Enable SDK or HTTP logging.
Inspect the REST endpoint and api-version being used.
Compare those API versions against the published retirement notices.

That will tell you definitively which applications require changes, regardless of the SDK package name.

Please "Accept the Answer" if this information helped you. This will help us and others in the community as well.

Luke Feetterer 0 Reputation points

2026-08-06T16:46:24.15+00:00

@Allan Solomon Mejia Thank you for your help and for the extra info, it is much appreciated.

One last nuance / clarification around the questionable usage of the Assistants API through Azure.AI.Projects or Azure.AI.Agents.Persistent.

If we're using threads / vector stores / runs but with the PersistentAgentsClient, is this the same thing as using assistants under the hood (and this code is still at risk)? This is where I'm getting lost, because the method signatures for this client still use "assistantId", etc. verbiage, and our "agent IDs" are still the "asst_..." values.
Allan Solomon Mejia 7,995 Reputation points

2026-08-06T16:54:57.3933333+00:00
Hi @Luke Feetterer

Thanks for the follow-up. This is an important distinction.

PersistentAgentsClient is part of the Azure AI Agents SDK, but it still interoperates with the same underlying persistent agent resources (which is why you'll continue to see concepts like assistantId, thread, run, and resource IDs beginning with asst_...). Those identifiers are implementation details and don't necessarily indicate you're calling the legacy Assistants API directly.

The key question is which REST endpoint and API version your application is actually invoking.

If your requests are going to the newer Azure AI Agents API endpoints with a supported API version, you're aligned with the current platform, even if the SDK still exposes familiar terminology.

If the SDK is issuing requests to the legacy Assistants API endpoints (or an API version that's being retired), then those calls would indeed be subject to the retirement timeline.

The most reliable way to verify this is to enable HTTP/SDK logging (or use a tool such as Fiddler, mitmproxy, or your application's diagnostics) and inspect:

The request URL/endpoint

The api-version query parameter

That will definitively tell you whether you're using the new Agents service or the legacy Assistants implementation under the hood.

If you can share:

the version of Azure.AI.Agents.Persistent (or Azure.AI.Projects) you're using, and

a sample HTTP request URL from the SDK logs (you can redact sensitive information),

I can help determine whether your application is affected by the Assistants API retirement.

Please "Accept the Answer" if this information helped you. This will help us and others in the community as well.
Allan Solomon Mejia 7,995 Reputation points

2026-09-02T18:37:51.47+00:00

Hello @Luke Feetterer

Help make this community better for everyone: if this answer resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution.