Intermittent 404 "Resource not found" on /openai/v1/decisions (gpt-6-luna) while chat completions on the same deployment work

Choi, SeungWan 0 Reputation points
2026-11-17T04:03:53.87+00:00

We call the Decisions API from a production app. Several times a day, every request to the Decisions endpoint returns 404 for 10+ minutes, and then it recovers by itself.

  • Endpoint: POST https://<my-resource>.openai.azure.com/openai/v1/decisions
  • Region: Korea Central. Deployment: gpt-6-luna (Global Standard)
  • Response: 404 {"error":{"code":"404","message":"Resource not found"}}
    • The 404 returns in ~20–50 ms with no x-ms-region header.
    • Successful calls take ~700 ms and return x-ms-region: Korea Central.

What we ruled out

  • Resource, key, and deployment. During an outage (2026-11-17 03:36 UTC), chat completions on the same resource, key, and gpt-6-luna deployment returned 200. This held for both /openai/deployments/.../chat/completions and /openai/v1/chat/completions. /openai/v1/decisions returned 404 at the same moment.
  • Client and auth. The 404s happen with an api-key header, with an Authorization: Bearer header, and with openai-python 3.51.0 client.decisions.create().
  • Input and api-version. Different inputs fail the same way. Adding api-version doesn't help: it still returns 404 during an outage and 400 "API version not supported" when the service is healthy.

Outage windows (UTC)

  • 2026-11-16: 01:52, 02:40, 03:50–03:53, 09:03–09:05, 10:04–10:06
  • 2026-11-17: ~03:24–03:37

Sample apim-request-id values

2026-11-16 01:52:09Z 0990f2c8-097f-4704-ac52-5e8e1c915a56

2026-11-16 02:40:07Z ffe8fbc5-097f-4e5e-8dd8-6c759067168a

2026-11-16 03:52:06Z 9b586b0e-2651-4776-ad34-c8871805a8c4

2026-11-16 10:04:57Z 2cb4a204-4efe-44da-a2e1-272e1c40f445

2026-11-16 10:06:01Z ca5b357e-f660-49c4-bb07-6111951f6648

Questions

1. What causes these 404s on the Decisions endpoint?

2. The docs say Decisions is "rolling out to regions where gpt-6-luna is available". With our deployment type, can requests be routed to a backend where Decisions isn't available yet?

3. Is there a recommended region or deployment type to avoid this?

Azure OpenAI in Foundry Models
0 comments No comments

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.