Seeing 503 service not available for GPT 5.6 sol model on foundry for Oct 21

Question

Seeing 503 service not available for GPT 5.6 sol model on foundry for Oct 21

Das, Sumeet (Contractor) 20

Continuously seeing 503 when trying to reach the GPT model 5.6 and in the logs seeing this error :
was the model down yesterday

fastapi.exceptions.HTTPException: 504: Could not call azure openai: {
  "error": {
    "message": "The server had an error while processing your request. Sorry about that!",
    "type": "server_error",
    "param": null,
    "code": null
  }
}

Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-22T10:27:08.16+00:00
Hi @Das, Sumeet (Contractor)

Thank you for reaching out on Microsoft Q&A.

Please follow the steps in the AI-recommended answer. A few additional points:

Confirm the timeline from your own metrics: In the Azure portal, open your resource → Monitoring → Metrics → Azure OpenAI Requests. Apply splitting by StatusCode and filter by ModelDeploymentName. This shows exactly when the 5xx errors started and whether they're still happening.

Find out whether it was a platform event: If Service Health → Health history shows an event for your region on October 21, open it. Microsoft publishes the root cause and Post Incident Review (PIR) there, which answers whether the model was down.

Reduce regional exposure: If your deployment type is Standard (regional), consider a Global Standard or Data Zone Standard deployment. These route traffic across more capacity and are usually more resilient to capacity issues in one region. Please check that the model and deployment type are available in your region first.

If the issue persists, please open an Azure support request from your resource: Support + Troubleshooting → Create a support request. Include the region, deployment name, model version, API version, UTC timestamps of the failures and the apim-request-id values from failed responses. Support can then check backend telemetry for your deployment.

If this helps, please click Accept Answer and Upvote. Thank you!

For the Translator case: remove the private message section. In the posted comment, replace "reply here and we'll follow up privately" with: "Please open an Azure support request from the resource (Support + Troubleshooting → Create a support request) so the backend team can check the global gateway mapping."

Thankyou
Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-23T08:09:59.3466667+00:00

Hi @Das, Sumeet (Contractor) Did you get any chance to review the above response. Do let me know if you have any further queries.
Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-26T09:15:53.03+00:00

Hi @Das, Sumeet (Contractor) We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.
Das, Sumeet (Contractor) 20 Reputation points

2026-10-26T12:16:53.7166667+00:00

Thanks for the reply
It helped!!

Answer accepted by question author

0 additional answers

Your answer

Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-22T10:27:08.16+00:00

Hi @Das, Sumeet (Contractor)

Thank you for reaching out on Microsoft Q&A.

Please follow the steps in the AI-recommended answer. A few additional points:

Confirm the timeline from your own metrics: In the Azure portal, open your resource → Monitoring → Metrics → Azure OpenAI Requests. Apply splitting by StatusCode and filter by ModelDeploymentName. This shows exactly when the 5xx errors started and whether they're still happening.

Find out whether it was a platform event: If Service Health → Health history shows an event for your region on October 21, open it. Microsoft publishes the root cause and Post Incident Review (PIR) there, which answers whether the model was down.

Reduce regional exposure: If your deployment type is Standard (regional), consider a Global Standard or Data Zone Standard deployment. These route traffic across more capacity and are usually more resilient to capacity issues in one region. Please check that the model and deployment type are available in your region first.

If the issue persists, please open an Azure support request from your resource: Support + Troubleshooting → Create a support request. Include the region, deployment name, model version, API version, UTC timestamps of the failures and the apim-request-id values from failed responses. Support can then check backend telemetry for your deployment.

If this helps, please click Accept Answer and Upvote. Thank you!

For the Translator case: remove the private message section. In the posted comment, replace "reply here and we'll follow up privately" with: "Please open an Azure support request from the resource (Support + Troubleshooting → Create a support request) so the backend team can check the global gateway mapping."

Thankyou
Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-23T08:09:59.3466667+00:00

Hi @Das, Sumeet (Contractor) Did you get any chance to review the above response. Do let me know if you have any further queries.
Shaik Haaris 5 Reputation points Microsoft External Staff

2026-10-26T09:15:53.03+00:00

Hi @Das, Sumeet (Contractor) We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.
Das, Sumeet (Contractor) 20 Reputation points

2026-10-26T12:16:53.7166667+00:00

Thanks for the reply
It helped!!

Answer 1

A broad GPT-5.6 outage on October 21, 2026 cannot be confirmed from the public Azure status history; check Azure Service Health and Resource Health for the affected subscription, resource, and deployment region.

The model response is a server-side server_error. Microsoft classifies HTTP 500/503 responses as transient service issues or backend overload, with retry, service-health checking, and support escalation as the documented actions.^1^ The 504 shown by FastAPI appears to be the status generated by your application after the Azure OpenAI call failed; the embedded Azure response does not itself establish a platform-wide outage.

Check whether your deployment was affected

In the Azure portal, search for Service Health.
Select Service issues and set the time range to October 21.
Filter for the subscription and deployment region, and include:
- Azure OpenAI
- Foundry Models
Open Resource Health for the specific Foundry/Azure OpenAI resource and review recent unavailability.

Service Health is subscription- and region-specific, so an event may appear there without being listed on the broad public Azure status page. Microsoft recommends Service Health and Resource Health to determine the scope of a rise in 5xx errors.^2^ The public status page does not provide clear confirmation of a GPT-5.6 incident for that date.^3^

If requests are still failing

Test the same deployment directly in the Foundry playground.
- If both the playground and API return 503, the problem is downstream of your FastAPI application.
- If only FastAPI fails, inspect its timeout and exception mapping.
Record:
- Exact UTC timestamps
- Azure region
- Resource and deployment names
- Model name and version as displayed in Foundry
- API version
- Request IDs and response headers
- Whether every request or only some requests fail
Retry transient 500/503 responses using exponential backoff with jitter. Do not retry indefinitely.
If failures continue for more than a few minutes, open an Azure support request and include the collected request IDs.^1^

For production continuity, switch to a pre-created deployment in a secondary region. Microsoft recommends regional failover for sustained 5xx errors because API request processing still has regional dependencies.^2^

Also confirm the exact model identifier: the question says GPT 5.6 sol, while a related report names GPT-5.6-Luna. The region and exact deployment/model version are required to determine whether both reports concern the same service event.[^4^](https://learn.microsoft.com/en-us/answers/questions/5989897/gpt-5-6-luna-deployment-returning-503-internalserv "GPT-5.6-Luna deployment returning 503 InternalServerError: "no healthy ...")

References