Transcription deployments return DeploymentNotFound; new deployment stuck with RequestConflict

We operate these models in swedencentral. There were no previous issues with the deployments, they suddenly stopped working.

Problem 1: Existing deployments are not served by the data plane.

The deployments managed-gpt-transcribe (model gpt-transcribe, version 2026-07-28, Global Standard) and managed-gpt-4o-mini-transcribe (model gpt-4o-mini-transcribe, version 2025-12-15, Global Standard) both show provisioning state "Succeeded" in the portal. Every transcription request to them returns HTTP 404 DeploymentNotFound:

POST https://xxx.openai.azure.com/openai/deployments/managed-gpt-transcribe/audio/transcriptions?api-version=2025-03-01-preview → 404 {"error":{"code":"DeploymentNotFound","message":"The API deployment for this resource does not exist..."}}

We get the same result with several api-versions (2024-06-01, 2024-10-21, 2025-01-01-preview, 2025-03-01-preview, 2025-04-01-preview), the /openai/v1/audio/transcriptions path, and the cognitiveservices.azure.com and services.ai.azure.com hostnames. The API key is valid: a wrong key returns 401, and other deployments on the same resource (e.g. managed-gpt-realtime-2.1-mini) resolve normally. The portal offers no playground for these deployments.

Problem 2: Redeploying fails.

When we try to redeploy gpt-transcribe, the first Write Deployment failed with GatewayTimeout (504):

• Correlation ID: 03555b11-cf17-4af0-ab44-ea352ba1205f
• Timestamp: 2026-09-29T13:11:52Z
• Message: "The gateway did not receive a response from 'Microsoft.CognitiveServices' within the specified time period."

One Write Deployment operation has stayed in "Started" state since then. All further attempts fail with:

• RequestConflict: "Another operation is being performed on the parent resource"
• Trace ID: bf5bde5a-3cb5-4822-83da-24d60ce1f6d6
• Service request ID: 4bf91de3-faaf-40bc-9f93-7c305979aca0