After an Azure OpenAI model version is retired, Microsoft says every inference request returns 410 Gone. Several o-series versions have retirement date November 19, 2026 on the schedule and in SunsetDiff.
Microsoft's lifecycle page does not print an error sentence. It says a Retired model is removed from service and all inference requests return 410 Gone.
The lifecycle table says Retired means removed from service, new deployments are not allowed, existing deployments do not work, and all inference requests return 410 Gone. The same page says the API value lifecycleStatus Deprecated means the model is retired and inference returns 410 Gone. The docs stage Deprecated is lifecycleStatus Deprecating, and inference still works. The retirement schedule, checked October 24, 2026, lists these Azure OpenAI base rows as Deprecated with retirement date 2026-11-19: o1 2024-12-17 and o1-pro 2025-03-19 and o3 2025-04-16 and o3-deep-research 2025-06-26 and o3-pro 2025-06-10, replacement gpt-5.6-sol; o3-mini 2025-01-31, replacement gpt-5.6-terra. SunsetDiff stores shutdown_date 2026-11-19 for those six ids. The schedule's base row for o4-mini 2025-04-16 is also Deprecated, retirement 2026-11-19, replacement gpt-5.6-terra. SunsetDiff's file for o4-mini-2025-04-16 stores shutdown_date 2027-11-19, which is the fine-tuned deployment retirement on that schedule, so this page does not treat the stored date as the base-row date. The lifecycle page says Global Standard, Data Zone Standard, and Standard deployments are auto-upgraded at retirement, and provisioned deployments are not. The working-with-models page says automatic model updates are supported for Standard deployment types. On that page, NoAutoUpgrade never upgrades and the deployment stops working at retirement. OnceNewDefaultVersionAvailable upgrades within two weeks of a new default. OnceCurrentVersionExpired upgrades when the retirement date is reached. Source: Foundry Models lifecycle, checked October 24, 2026.
| Model id | What the source says |
|---|---|
o1-2024-12-17 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-sol. |
o1-pro-2025-03-19 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-sol. |
o3-2025-04-16 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-sol. |
o3-mini-2025-01-31 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-terra. |
o3-deep-research-2025-06-26 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-sol. |
o3-pro-2025-06-10 | Schedule and SunsetDiff: Deprecated, retirement 2026-11-19, replacement gpt-5.6-sol. |
o4-mini-2025-04-16 | The schedule's base row says Deprecated, retirement 2026-11-19, replacement gpt-5.6-terra. SunsetDiff stores shutdown_date 2027-11-19 from the fine-tuned deployment row. |
gpt-5.6-sol | Replacement named for o1, o1-pro, o3, o3-deep-research, and o3-pro. |
gpt-5.6-terra | Replacement named for o3-mini. The schedule also names it for the o4-mini base row. |
Point the deployment at the replacement in the table before the retirement date. Working with models says automatic updates are supported for Standard deployment types. NoAutoUpgrade never upgrades, and at retirement that deployment stops working. You then point the code at a deployment that has not expired. The replacement ids are the ones in the table, from the retirement schedule.
Source: Working with Azure OpenAI models, checked October 24, 2026.
This page does not show a LiteLLM call. Change the Azure deployment the proxy uses so it is a deployment of the replacement id on the schedule. The 410 status is on the lifecycle page, after the model is retired.
Source: Foundry Models lifecycle, checked October 24, 2026.
This page does not show an AI SDK call. The Azure target is a deployment of the replacement model. The lifecycle page is the source for 410 Gone.
Source: Azure OpenAI retirement schedule, checked October 24, 2026.
The schedule lists them as Deprecated, which the lifecycle page maps to API status Deprecating. Inference still works in that stage. 410 Gone is what the page says a Retired model returns. The retirement date on the matching SunsetDiff rows is 2026-11-19.
The lifecycle page says Global Standard, Data Zone Standard, and Standard deployments are auto-upgraded at retirement, and provisioned deployments are not. Working with models says automatic updates are supported for Standard deployment types. NoAutoUpgrade never upgrades, and that deployment stops working at retirement.
The stored shutdown_date is the fine-tuned deployment retirement. The schedule's base row for o4-mini 2025-04-16 says retirement 2026-11-19 and replacement gpt-5.6-terra. Read both on o4-mini 2025-04-16.