Hi there
This can happen occasionally when the request hits a backend routing node that hasn’t refreshed deployment metadata yet. If the exact same call works on retry, it’s not a config issue on your side — it’s a transient gateway/region lookup glitch. The best practice is to add lightweight retry logic (e.g., 1–2 retries with exponential back-off) since Azure OpenAI can briefly return these errors during scale or cache refresh events. Also double-check you're using the regional endpoint and not the global one, as global routing can make this happen a bit more often.
If this helps kindly accept the answer.