Azure OpenAI gpt-4 preview deprecated — 404 on retired model version

"""Audit all Azure OpenAI deployments for preview / deprecated model versions."""
from azure.identity import DefaultAzureCredential
from azure.mgmt.resource import SubscriptionClient, ResourceManagementClient
from azure.mgmt.cognitiveservices import CognitiveServicesManagementClient

# Model versions that are preview or deprecated as of 2026
RETIRED_OR_DEPRECATED = {
    "gpt-4-vision-preview": "retired 2024-08",
    "gpt-4-1106-preview":   "retired 2025-04",
    "gpt-4-0125-preview":   "retired 2025-04",
    "gpt-4-turbo-2024-04-09": "deprecated — retire Q3 2025",
    "gpt-35-turbo-1106":    "deprecated — retire Q2 2025",
    "gpt-35-turbo-instruct": "deprecated — retire 2026",
}

credential = DefaultAzureCredential()
sub_client = SubscriptionClient(credential)

print(f"{'Subscription':<40} {'Resource':<30} {'Deployment':<30} {'Model':<25} {'Version':<15} {'Status':<25}")
print("-" * 165)

for sub in sub_client.subscriptions.list():
    sub_id = sub.subscription_id
    sub_name = sub.display_name
    cog_client = CognitiveServicesManagementClient(credential, sub_id)
    try:
        accounts = list(cog_client.accounts.list())
    except Exception as e:
        print(f"{sub_name:<40} (error listing accounts: {e})")
        continue

    for account in accounts:
        if account.kind != "OpenAI":
            continue
        rg = account.id.split("/")[4]
        for deployment in cog_client.deployments.list(rg, account.name):
            model = deployment.properties.model
            key = f"{model.name}-{model.version}" if model.version else model.name
            issue = RETIRED_OR_DEPRECATED.get(model.name, "")
            if not issue and "preview" in (model.version or "").lower():
                issue = "preview version — verify GA status"
            status = issue or "OK"
            print(f"{sub_name:<40} {account.name:<30} {deployment.name:<30} {model.name:<25} "
                  f"{model.version or '-':<15} {status:<25}")

RG=my-rg
AOAI=my-aoai-resource
DEPLOYMENT=gpt4-prod

# 1) Verify current model
az cognitiveservices account deployment show \
  --name $AOAI -g $RG --deployment-name $DEPLOYMENT \
  --query "{model:properties.model.name, version:properties.model.version, sku:sku.name}"

# 2) Update to gpt-4o (GA)
az cognitiveservices account deployment update \
  --name $AOAI -g $RG --deployment-name $DEPLOYMENT \
  --model-name gpt-4o \
  --model-version 2024-08-06 \
  --model-format OpenAI

# 3) Verify update succeeded
az cognitiveservices account deployment show \
  --name $AOAI -g $RG --deployment-name $DEPLOYMENT \
  --query "properties.model"

# 4) Test the new model with a probe request
curl -X POST "https://$AOAI.openai.azure.com/openai/deployments/$DEPLOYMENT/chat/completions?api-version=2024-10-21" \
  -H "Content-Type: application/json" \
  -H "api-key: $AOAI_KEY" \
  -d '{"messages":[{"role":"user","content":"OK"}],"max_tokens":10}'

"""Middleware to detect deprecation warning headers on every AOAI response."""
import logging
import httpx
from openai import AzureOpenAI

log = logging.getLogger(__name__)

class DeprecationWarningMiddleware(httpx.Client):
    """Wraps the OpenAI SDK's HTTP client to inspect deprecation headers."""

    def send(self, request, **kwargs):
        response = super().send(request, **kwargs)
        deprecation = response.headers.get("x-ms-model-deprecation-notice")
        retire_date = response.headers.get("x-ms-model-retirement-date")
        substituted = response.headers.get("x-ms-deployment-model-substituted")

        if deprecation or retire_date:
            log.warning(
                "AOAI deprecation warning: %s (retire: %s) on %s",
                deprecation, retire_date, request.url,
            )
            # Emit metric for alerting
            emit_metric("aoai.deprecation_warning", 1, tags={
                "retire_date": retire_date or "unknown",
            })

        if substituted == "true":
            log.error(
                "AOAI model auto-substituted: %s -> %s. Response quality may have changed.",
                response.headers.get("x-ms-original-model"),
                response.headers.get("x-ms-substituted-model"),
            )
            emit_metric("aoai.model_substituted", 1)

        return response

def emit_metric(name, value, tags=None):
    """Send to your telemetry backend."""
    pass

# Wire the middleware into the OpenAI client
client = AzureOpenAI(
    api_key="...",
    azure_endpoint="...",
    api_version="2024-10-21",
    http_client=DeprecationWarningMiddleware(),
)