Subject: GA-status models (gpt-4o 2024-11-20 / gpt-4o-mini 2024-07-18) are blocked for new deployments — this directly contradicts Microsoft's own documentation

We are unable to create new deployments of gpt-4o (2024-11-20) or gpt-4o-mini (2024-07-18) in any capacity, and the error returned directly contradicts Microsoft's own published lifecycle documentation. We need clarification, because this is impacting a production workload and our migration planning.

Here is the contradiction laid out precisely, with references to the official docs.

---

1. The blocked models are documented as GA — not Deprecated.

Per the official Model retirement schedule (https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule?view=foundry-classic), the exact versions we are trying to deploy are listed as:

Model	Version	Lifecycle	Retirement date
gpt-4o	2024-11-20	GA	2026-10-01
gpt-4o	2024-11-20	GA	2026-10-01
gpt-4o-mini	2024-07-18	GA	2026-10-01

These are GA, not "Deprecated" and not "Legacy." Their retirement date is October 2026 — nearly 3 months away.

---

2. Microsoft's lifecycle policy states GA models can be deployed by new customers.

Per the Model lifecycle and support policy (https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements?view=foundry-classic), the lifecycle stage table explicitly states:

• Generally Available (GA) → "Can create new deployments? Yes"
• Only the Deprecated stage blocks new customers, and that stage begins "at 12 months from launch."

By Microsoft's own rules, a GA model must be deployable by new customers until it transitions to Deprecated. These versions have not transitioned — the schedule still lists them as GA.

---

3. The error we receive maps to the Deprecated stage — not GA.

The lifecycle documentation contains this API-to-portal terminology mapping:

API lifecycleStatus: "Deprecating" = portal/docs stage "Deprecated" = "Still serves inference. Blocked for new customers."

The error returned is:

ServiceModelDeprecating: The model 'Format:OpenAI,Name:gpt-4o,Version:2024-11-20' is in deprecating state and cannot be used for new deployments.

This means the backend is reporting these versions as Deprecated (blocked for new customers) — while the public retirement schedule simultaneously lists them as GA.

Both states cannot be true at once. Either:

• (a) the backend is incorrectly flagging GA models as Deprecated — a bug; or
• (b) the documentation is wrong and has not been updated to reflect the real state.

Either way, customers are making production decisions based on a GA status that does not reflect reality.

---

4. If this is regional "new customer" limiting, it needs to be stated transparently.

The lifecycle policy includes the clause: "Microsoft can limit new customers in specific regions to maintain service quality for existing customers."

If that clause is being applied here, it is not reflected anywhere in the GA status shown in the retirement schedule. Showing a model as "GA" while silently blocking new deployments in a region is not acceptable transparency for a production service.

---

5. The documented replacement chain is itself broken.

The retirement schedule lists the replacement for gpt-4o-mini as gpt-4.1-mini — but gpt-4.1-mini is also marked Deprecated (retires 2026-10-14) and is also blocked with the same ServiceModelDeprecating error:

ServiceModelDeprecating: The model 'Format:OpenAI,Name:gpt-4.1-mini,Version:2025-04-14' is in deprecating state and cannot be used for new deployments.

Recommending a deprecated, non-deployable model as the official replacement for another model makes it impossible to plan a migration based on the schedule.

---

What we need clarified:

1. Are gpt-4o (2024-11-20) and gpt-4o-mini (2024-07-18) being treated as Deprecated (existing-customers-only) at the backend, despite being published as GA?
2. If yes — under what clause, and why is the public schedule not updated to reflect it?
3. If this is "new customer limiting in specific regions," please confirm it in writing and state which regions are affected.
4. Is there any path for a subscription to gain new-deployment access to these GA-listed versions, or has access been revoked ahead of the documented retirement date?
5. When will the schedule be corrected so it no longer shows misleading GA status for versions that cannot actually be deployed?
6. Given that the listed replacement (gpt-4.1-mini) is itself deprecated and blocked, what is the actual deployable replacement at a comparable price point?

This is affecting a live production workload. We would appreciate this being escalated to the relevant product team if the GA-vs-Deprecated discrepancy cannot be resolved directly.

Thank you.We are unable to create new deployments of gpt-4o (2024-11-20) or gpt-4o-mini (2024-07-18) in any capacity, and the error returned directly contradicts Microsoft's own published lifecycle documentation. We need clarification, because this is impacting a production workload and our migration planning.

Here is the contradiction laid out precisely, with references to the official docs.

---

1. The blocked models are documented as GA — not Deprecated.

Per the official Model retirement schedule (https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule?view=foundry-classic), the exact versions we are trying to deploy are listed as:

Model	Version	Lifecycle	Retirement date
gpt-4o	2024-11-20	GA	2026-10-01
gpt-4o-mini	2024-07-18	GA	2026-10-01

These are GA, not "Deprecated" and not "Legacy." Their retirement date is October 2026 — nearly a year away.

---

2. Microsoft's lifecycle policy states GA models can be deployed by new customers.

Per the Model lifecycle and support policy (https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements?view=foundry-classic), the lifecycle stage table explicitly states:

• Generally Available (GA) → "Can create new deployments? Yes"
• Only the Deprecated stage blocks new customers, and that stage begins "at 12 months from launch."

By Microsoft's own rules, a GA model must be deployable by new customers until it transitions to Deprecated. These versions have not transitioned — the schedule still lists them as GA.

---

3. The error we receive maps to the Deprecated stage — not GA.

The lifecycle documentation contains this API-to-portal terminology mapping:

API lifecycleStatus: "Deprecating" = portal/docs stage "Deprecated" = "Still serves inference. Blocked for new customers."

The error returned is:

ServiceModelDeprecating: The model 'Format:OpenAI,Name:gpt-4o,Version:2024-11-20' is in deprecating state and cannot be used for new deployments.

This means the backend is reporting these versions as Deprecated (blocked for new customers) — while the public retirement schedule simultaneously lists them as GA.

Both states cannot be true at once. Either:

• (a) the backend is incorrectly flagging GA models as Deprecated — a bug; or
• (b) the documentation is wrong and has not been updated to reflect the real state.

Either way, customers are making production decisions based on a GA status that does not reflect reality.

---

4. If this is regional "new customer" limiting, it needs to be stated transparently.

The lifecycle policy includes the clause: "Microsoft can limit new customers in specific regions to maintain service quality for existing customers."

If that clause is being applied here, it is not reflected anywhere in the GA status shown in the retirement schedule. Showing a model as "GA" while silently blocking new deployments in a region is not acceptable transparency for a production service.

---

5. The documented replacement chain is itself broken.

The retirement schedule lists the replacement for gpt-4o-mini as gpt-4.1-mini — but gpt-4.1-mini is also marked Deprecated (retires 2026-10-14) and is also blocked with the same ServiceModelDeprecating error:

ServiceModelDeprecating: The model 'Format:OpenAI,Name:gpt-4.1-mini,Version:2025-04-14' is in deprecating state and cannot be used for new deployments.

Recommending a deprecated, non-deployable model as the official replacement for another model makes it impossible to plan a migration based on the schedule.

---

What we need clarified:

1. Are gpt-4o (2024-11-20) and gpt-4o-mini (2024-07-18) being treated as Deprecated (existing-customers-only) at the backend, despite being published as GA?
2. If yes — under what clause, and why is the public schedule not updated to reflect it?
3. If this is "new customer limiting in specific regions," please confirm it in writing and state which regions are affected.
4. Is there any path for a subscription to gain new-deployment access to these GA-listed versions, or has access been revoked ahead of the documented retirement date?
5. When will the schedule be corrected so it no longer shows misleading GA status for versions that cannot actually be deployed?
6. Given that the listed replacement (gpt-4.1-mini) is itself deprecated and blocked, what is the actual deployable replacement at a comparable price point?

This is affecting a live production workload. We would appreciate this being escalated to the relevant product team if the GA-vs-Deprecated discrepancy cannot be resolved directly.

Thank you.