CLU Model Expiry Date

Question

CLU Model Expiry Date

Yin, Sue 20

Hi team! I have a few questions related to the CLU Conversational Language Understanding Models.

I have created a few projects under "Language Studio" since 2023. While reviewing them recently, I found there is a "Model Expiry Date is 2026/2/27" for all of models I created. Does it mean these models will all be retired from 2026/2/27? But at the same time, I see "Deployment Expiry Date is 2027/2/27" for my trained deployments as well. How should I understand for these two dates?

Beside, while I am walking through the docs, there is no communication for if CLU will be retired in 2026. But from here "https://learn.microsoft.com/en-us/azure/ai-services/language-service/concepts/model-lifecycle", I found that for "Conversational language understanding", the Training Config Expiration August 26, 2025 and Deployment Expiration is August 26, 2026 for latest version 2022-09-01. How should I understand this? And will we have new version of training config?

If CLU will be retired soon, what should we do to migrate our current models?

Yin, Sue 20 Reputation points

2025-11-20T07:16:14.1233333+00:00

Since the latest Supported Training Config Versions is 2022-09-01, which will be expired with above dates on 2026-08-26, what's the new version should I use?

Answer accepted by question author

1 additional answer

Your answer

Yin, Sue 20 Reputation points

2025-11-20T07:16:14.1233333+00:00

Since the latest Supported Training Config Versions is 2022-09-01, which will be expired with above dates on 2026-08-26, what's the new version should I use?

Answer 1

Hello Yin, Sue,

Welcome to Microsoft Q&A,

Thanks for raising this the expiry dates shown in Language Studio can definitely be confusing.

Let me break down what each date means and how it aligns with the official CLU model lifecycle.

The “Model Expiry Date” (for example, 2026-02-27) refers to the training configuration version your model was originally trained with. After this date, you will no longer be able to retrain that model or create new models using that same training config. The model can still run predictions until the deployment expires, but retraining or updating it will not be supported.

The “Deployment Expiry Date” (for example, 2027-02-27) is the date until which your deployed endpoint continues to function normally. Once this date passes, the deployment will stop serving predictions unless you redeploy using a supported training configuration. This means the deployment stays valid for longer, even though the underlying model’s training config has already expired.

The dates you found in Microsoft’s lifecycle documentation apply to the training configuration version (2022-09-01), not to individual models. The documentation lists a Training Config Expiration of August 26, 2025, and a Deployment Expiration of August 26, 2026. The dates shown in your Language Studio UI are calculated based on when your model was trained, so they may differ slightly but still fall within the official lifecycle window.

Microsoft has not announced retirement for CLU itself. The only thing expiring is the current training configuration. A new training config version is expected around June 2025, with improvements for multilingual scenarios and reduced over-predictions. Once this new version becomes available, it will replace the current 2022-09-01 configuration.

To prepare, you can continue using your current models normally until the existing training configuration expires on August 26, 2025. Once the new training config is released in 2025, you should retrain your models using that updated version and redeploy them before your deployment expiration window in 2026–2027. This ensures uninterrupted service with the latest supported configuration.

There is no requirement to migrate away from CLU only to move to the newer training configuration version once it becomes available.

Please refer this

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!

Yin, Sue 20 Reputation points

2025-11-21T02:33:13.4+00:00

Hi @SRILAKSHMI C , thank you for the details. That's really helpful!

I saw there is a new training version named "2025-07-01Preview". Is this the newest one? Although while selecting the training models, it shows "2022-09-01" is the latest one. This also makes me a bit confused as well...

I tried to retrain with "2025-07-01Preview" training model, then the model expiry date was refreshed to "2026-07-01". Is it expected?
SRILAKSHMI C 10,885 Reputation points Microsoft External Staff Moderator

2025-11-24T17:56:15.3733333+00:00

Hi Yin, Sue,

Thanks for checking this, the behavior you’re seeing is expected.

1. About the “2025-07-01Preview” training version

Yes, 2025-07-01Preview is the newest training configuration, but it is not yet the default. That’s why:

In the UI, the “latest” label still appears next to 2022-09-01

But in the dropdown, you can manually select 2025-07-01Preview

Preview versions are offered for early testing and evaluation before Microsoft promotes them to “latest” or “GA”.

2. Why the expiry date changed after retraining

This is also expected.

CLU expiry dates are tied to the training configuration version, not the project itself. So when you retrain using 2025-07-01Preview, the system recalculates:

Model Expiry Date → 12 months after the preview config’s publish date

Deployment Expiry Date → 24 months after the publish date

Since the preview config is dated 2025-07-01, you seeing:

Model Expiry: 2026-07-01 matches the standard lifecycle rule.

3. Why the UI still shows 2022-09-01 as “latest”

The UI only labels a training config as “latest” after it becomes GA (General Availability). Right now:

2022-09-01 → current GA version

2025-07-01Preview → new preview version, not GA yet

Once Microsoft finalizes the 2025 config, it will replace 2022-09-01 and become the new default.

You can Continue testing with the 2025-07-01Preview version

Or wait for the GA version before retraining production models

Thank you!
Yin, Sue 20 Reputation points

2025-11-25T02:12:20.8+00:00

Hi @SRILAKSHMI C , appreciate for all of the details, that helps a lot! Thank you!

Answer 2

The "Model Expiry Date" you see for your CLU models indicates that these models will no longer be available for use after that date, which is set for February 27, 2026. This means that you will need to retrain your models or migrate to a newer version before this date to ensure continued functionality.

The "Deployment Expiry Date" of February 27, 2027, refers to the expiration of the deployments that were created from these models. After the model expiry, the deployments will also become inactive if they are based on the expired models. Therefore, it is essential to manage both dates: retrain your models before the model expiry and ensure your deployments are updated accordingly.

Regarding the training configuration expiration, the information you found indicates that the latest training configuration version for Conversational Language Understanding (CLU) has an expiration date of August 26, 2025. After this date, you will need to use a newer training configuration version to submit training or deployment jobs. If no newer versions are available, the expiration date may be extended, but it is advisable to keep an eye on updates from Microsoft regarding new versions.

As for the retirement of CLU itself, there is no current indication from the documentation that CLU will be retired in 2026. However, it is always a good practice to stay updated with the official Microsoft announcements and documentation for any changes in service availability.

To migrate your current models, you should plan to retrain them using the latest supported training configuration version before the expiration dates. This will ensure that your models remain functional and up-to-date with any improvements made in the CLU service.

References:

Share via

CLU Model Expiry Date

1 additional answer

Your answer