Share via

Clarification needed: Future replacement for GPT-4o Standard deployments in Switzerland North

Michael Ammann 20 Reputation points
2026-05-22T06:39:50.02+00:00

We are currently running production workloads on Azure OpenAI in the Switzerland North region using GPT-4o with the Standard deployment type.

As GPT-4o retirement timelines are approaching, we are reviewing our migration options. At the moment, we are unable to identify a clearly equivalent successor model that is available for Standard deployments in Switzerland North.

For our planning, we would appreciate clarification on the expected path forward for customers in this setup.

Specifically, we would like to understand whether Microsoft plans to make a comparable replacement model available in Switzerland North under the Standard deployment type before GPT-4o is retired.

From a production perspective, the main points we need to assess are:

  • whether an in-region Standard deployment option will remain available after GPT-4o retirement
  • which model Microsoft recommends as the functional successor for current GPT-4o workloads in Switzerland North
  • whether the successor model is expected to offer comparable capabilities, latency, and quota behavior
  • whether customers will have an overlap period to test and validate the replacement model before GPT-4o is no longer available
  • whether migration to another deployment type, such as Global Standard, Data Zone, or Provisioned, should be expected if no Switzerland North Standard replacement is planned

Our concern is not only model retirement itself, but the lack of a visible like-for-like migration target in the same region and deployment type. Since this is already part of our production environment, we need to understand whether we can plan for a direct replacement or whether architectural changes will be required.

Could Microsoft please clarify the recommended migration strategy for GPT-4o Standard deployments in Switzerland North?

Thank you very much!

Azure OpenAI in Foundry Models

Answer recommended by moderator

Michael Ammann 20 Reputation points
2026-05-26T16:01:50.1633333+00:00

Thank you for the responses.

My question is not fully answered, but I understand that nobody can or wants to provide a binding commitment regarding future model availability in a specific region and deployment type.

A clearer roadmap for model availability in Swiss regions would be very helpful, especially for customers with data residency or resilience requirements. However, I also understand that this is difficult given the current pace at which new models are released and retired.

For now, I will most likely migrate from GPT-4o to GPT-4.1 to gain some additional time. After that, I may need to move to an EU region if no suitable successor becomes available in Switzerland North under the Standard deployment type. This would unfortunately mean losing Swiss data residency.

Thank you again for the clarification and the answers. I will close the case.

Was this answer helpful?

0 comments

4 additional answers

Sort by: Most helpful
Most helpful Newest Oldest
  1. kagiyama yutaka 3,420 Reputation points
    2026-05-22T09:43:35.58+00:00

    I’m not aware of any public Microsoft information confirming a successor model for Switzerland North / Standard, or any region‑ or SKU‑specific future availability. So at this point I don’t think anyone can give a definitive answer.

    Was this answer helpful?

    1 person found this answer helpful.
    0 comments
  2. Sina Salam 29,516 Reputation points Volunteer Moderator
    2026-05-26T17:15:24.0766667+00:00

    Hello Michael Ammann,

    Welcome to the Microsoft Q&A and thank you for posting your questions here.

    I understand that you need clarification on "Future replacement for GPT-4o Standard deployments in Switzerland North." and to my understanding you are asking if you “Can keep a Switzerland-based, Standard deployment path after GPT-4o retirement, or you must redesign” is not a generic “on the next model but a production architecture and compliance continuity question.

    You should not redesign away from Switzerland North Standard unless and until official controls prove there is no Standard replacement path at retirement time. The official Microsoft lifecycle already provides the mechanism you need:

    1. Treat gpt-5.1 as the official successor family for GPT-4o. That is the current replacement listed in the retirement schedule. - https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule
    2. Keep the workload on Switzerland North and Standard if Switzerland-only inference processing is mandatory. Standard is the deployment type that processes inference in the deployment region. Global Standard and Data Zone Standard do not preserve Switzerland-only processing. - https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/deployment-types, https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure-region-availability, https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy
    3. Set the deployment upgrade policy to OnceCurrentVersionExpired unless you have already validated the next default and want earlier auto-rollover. This gives you a controlled retirement-time upgrade path without changing application deployment names. Official model update policies support this. [- https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/working-with-models, https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/model-versions
    4. Do not wait for portal visibility alone. Use the official control sources below:
    5. Validate the replacement as soon as a test path exists, even if not in the same region initially. Microsoft only guarantees meaningful overlap globally, not necessarily in the same regional Standard SKU. - https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements
    6. If you require lower latency variance or guaranteed capacity and must remain Switzerland-only, the valid architectural fallback is Regional Provisioned in Switzerland North — not Global Standard and not Data Zone Standard. Provisioned is single-region but requires manual migration management. - https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/deployment-types, https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements

    I hope this is helpful! Do not hesitate to let me know if you have any other questions, steps or clarifications.

    Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.

    Was this answer helpful?

    0 comments
  3. Anshika Varshney 12,030 Reputation points Microsoft External Staff Moderator
    2026-05-22T08:50:14.74+00:00

    Hi Michael Ammann,

    Thanks for your question. This is a very important topic, especially when planning for long term deployments in Microsoft Foundry.

    Let me clarify this in a simple way

    First, about GPT 4o future GPT 4o is not guaranteed to stay as a long term model. Like all Azure AI Foundry models, it follows a lifecycle where models are eventually replaced by newer versions with better performance and capabilities. [learn.microsoft.com]

    This means over time, older models like GPT 4o will move to legacy, then deprecated, and finally retired.

    Now coming to your main concern about replacement

    As per current information, GPT 4o is already planned for retirement Some versions are scheduled to retire around March 2026 for standard deployments, and later in 2026 for other deployment types. [linkedin.com]

    Microsoft has also identified replacement models for GPT 4o. The primary replacement mentioned is GPT 5.1, which is a newer and more capable model family. [jinlee794.github.io]

    So yes, there is a transition path, but it is not a permanent guarantee that any single model like GPT 4o will remain unchanged forever.

    What this means for your application

    You should not design your solution assuming GPT 4o will always be available. Instead, plan for model upgrades as part of normal lifecycle.

    Here are some best practices.

    Use deployment names instead of hardcoding model version If you use a deployment name, you can update the model behind the scenes without changing application code.

    Regularly monitor model updates Azure introduces newer models with better performance, and you should evaluate and test them periodically.

    Plan migration early When a model enters deprecation phase, start testing the recommended replacement model to avoid last minute issues.

    Choose model based on use case. Newer models like GPT 5 series are designed for advanced reasoning and agent-based workflows, while other models like GPT 4.1 may still be useful for fast and high-volume scenarios. [learn.microsoft.com]

    In short: There is no guarantee that GPT 4o will remain available long term Yes, Microsoft provides replacement models such as GPT 5.1 You should design your solution to handle model evolution and upgrades

    Helpful reference documentation

    Model lifecycle and retirement guidance Microsoft Foundry model lifecycle

    Model catalog and latest availability Azure AI Foundry model catalog

    I hope this helps clarify things. Do let me know if you have any further queries.
    Thankyou!

    Was this answer helpful?

    0 comments

  4. AI answer

    2026-05-22T06:40:28.9133333+00:00

    For GPT-4o Standard deployments in Switzerland North, the following points are supported by current documentation and guidance:

    1. In-region Standard availability after GPT-4o retirement

    Model availability is managed centrally, model-by-model and region-by-region. Documentation explicitly states that:

    • Not all model version combinations will be available in all regions.
    • A GA model version N may upgrade to a future model version N+X in some regions based on capacity limitations, and the new version may not be separately available to test in that same region.

    There is no guarantee in the documentation that a specific successor model will be offered as a Standard deployment in a particular region (such as Switzerland North) after GPT-4o retirement. Regional availability can change based on capacity and service consolidation decisions.

    1. Recommended functional successor for GPT-4o

    The model retirement guidance states that when a model enters Legacy/Deprecated, customers should:

    • Consult the Model Retirement Schedule and use the “Suggested Replacement” column.
    • Follow the “Working with models” guidance to deploy, test, and migrate to the replacement.

    An explicit example is given where GPT-4o (2024-05-13 and 2024-08-06) Standard deployments were auto-upgraded to gpt-5.1 on the Standard SKU when those GPT-4o versions retired. That example shows the pattern Microsoft is using: GPT-4o → GPT-5.1 as the successor model family.

    However, the documentation does not commit that this exact mapping (or any specific successor) will be available as Standard in every region, including Switzerland North. The replacement and its availability are defined in the Model Retirement Schedule and by the centrally managed model availability matrix.

    1. Capabilities, latency, and quota behavior

    The model lifecycle and retirement documentation does not provide region-specific guarantees for:

    • Latency
    • Quota behavior
    • Exact capability parity

    It does state that customers should test applications with new models and versions, evaluate behavior, and update applications before retirement dates. This implies that capability and performance equivalence must be validated by customers as part of their migration planning.

    1. Overlap period for testing

    The model availability and lifecycle guidance includes:

    • At least one year of GA model availability in at least one region worldwide.
    • A 60-day window where a new GA model is available in at least one global or standard region before any upgrades happen.
    • For global deployments, future model versions starting with GPT-4o and GPT-4 0409 will be available alongside their successor (N and N+1) for comparison.

    Additionally, for government clouds, a 30-day overlap between versions is mentioned; for public cloud, the general rule is that customers have 60 days to try a new GA model in at least one region before upgrades.

    This establishes that there is an overlap period for testing, but not necessarily in the same region and deployment type. Model version N and N+1 might not both be available in the same region.

    1. Whether migration to another deployment type should be expected

    Key points from the lifecycle and deployment guidance:

    • Global Standard, Data Zone Standard, and Standard deployments can be auto-upgraded when a model version is retired.
    • Provisioned deployments are not auto-upgraded and must be migrated manually.
    • Model version N may upgrade to N+X in some regions even if N+X is not separately available in that region; customers can test N+X in other regions before upgrades.
    • Customers are advised to test new models and plan migrations based on the Model Retirement Schedule and model availability.

    The documentation does not mandate a move from Standard to Global Standard, Data Zone, or Provisioned for a specific region. However, it explicitly notes:

    • Model version N and N+1 may not always be available in the same region.
    • Some regions may not have all combinations of models and versions.

    This means that if a like-for-like Standard successor is not offered in Switzerland North, customers may need to consider:

    • Using another region where the successor model is available as Standard.
    • Using Global Standard or Data Zone deployments where appropriate.
    • Using Provisioned deployments if they need explicit control and are willing to manage upgrades manually.
    1. Recommended migration strategy for GPT-4o Standard in Switzerland North

    Based on the documented policies and examples, the recommended strategy is:

    1. Monitor the Model Retirement Schedule
      • Use the Model Retirement Schedule to identify the suggested replacement for the specific GPT-4o version in use and to see the retirement and upgrade timelines.
    2. Check model availability and lifecycle status
      • Use the Models API to check lifecycleStatus, deprecation, and deprecationDate for GPT-4o and its suggested replacement.
      • Verify which models and versions are currently available in Switzerland North Standard and in other regions.
    3. Test the suggested replacement model
      • Deploy the suggested replacement (for example, GPT-5.1 where available) in a region and deployment type where it is offered.
      • Evaluate functional behavior, latency, and quota characteristics for the workload.
    4. Plan for regional or deployment-type flexibility
      • If a Standard successor is not available in Switzerland North, plan for one of:
        • Using another region with Standard deployments of the successor model.
        • Using Global Standard or Data Zone Standard deployments if data residency and compliance requirements allow.
        • Using Provisioned deployments if more control over upgrades is required.
    5. Migrate before retirement
      • Update applications to use the new model and deployment configuration before the GPT-4o retirement date, following the “How to upgrade to a new model or version” guidance.

    The documentation does not provide a Switzerland North–specific commitment for a Standard successor to GPT-4o. Migration planning should therefore assume that regional availability may change and should be based on the Model Retirement Schedule, model availability tables, and testing of the suggested replacement model in the regions and deployment types where it is available.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.