Azure OpenAI - `model` parameter in request body causes failures when deployment name differs from model name

[gregsmi]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

When using AzureOpenAI client, the SDK sends the model parameter in the request body even though Azure OpenAI uses URL-based routing (/deployments/{deployment-id}/...). This causes failures when the deployment name doesn't match the model name exactly.

This is particularly problematic for gpt-image-1.5 because:

1. Azure resource naming rules prohibit periods (dots) in deployment names
2. The model name contains a dot: gpt-image-1.5
3. Therefore, deployment name can never match model name (e.g., must use gpt-image-1-5)

The Azure backend validates the body's model field against actual model names, rejecting valid deployment names.

To Reproduce

Steps to Reproduce

from openai import AzureOpenAI

client = AzureOpenAI(
    azure_endpoint="https://<resource>.openai.azure.com/",
    api_key="<key>",
    api_version="2024-10-21"
)

# Deployment name: "gpt-image-1-5" (dashes, due to Azure naming rules)
# Model name: "gpt-image-1.5" (dot)

response = client.images.generate(
    model="gpt-image-1-5",  # deployment name
    prompt="A sunset over mountains",
    size="1024x1024",
    n=1
)

---

Expected Behavior

Request succeeds because URL routing correctly targets the deployment:

POST /openai/deployments/gpt-image-1-5/images/generations

---

Actual Behavior

Request fails with:

{
    "error": {
        "message": "Model not supported with Responses API. Supported models are: ['gpt-image-1', 'gpt-image-1-mini', 'gpt-image-1.5']",
        "type": "invalid_request_error"
    }
}

The Azure backend is validating "model": "gpt-image-1-5" in the request body against model names, not deployment names.

---

Root Cause

In openai/lib/azure.py, the _build_request method uses model from the body to construct the URL but doesn't remove it from the body:

def _build_request(self, options, *, retries_taken=0):
    if options.url in _deployments_endpoints and is_mapping(options.json_data):
        model = options.json_data.get("model")
        if model is not None and "/deployments" not in str(self.base_url.path):
            options.url = f"/deployments/{model}{options.url}"
            # BUG: model is still in options.json_data

    return super()._build_request(options, retries_taken=retries_taken)

The Azure OpenAI REST API spec does not include model in the request body for deployment-based endpoints - model selection is done entirely via the URL path.

Code snippets

OS

Linux (Ubuntu)

Python version

3.12

Library version

2.7.1 (also tested against latest)

[xXMrNidaXx]

Confirmed — Great Catch!

This is indeed a bug in the Azure-specific request handling. The root cause is exactly as you identified.

Why This Happens

Looking at openai/lib/azure.py, the _build_request method uses the model field to construct the deployment URL path but doesn't strip it from the JSON body:

# Pseudocode of current behavior:
def _build_request(self, options, ...):
    model = options.json_data.get("model")
    if model is not None:
        options.url = f"/deployments/{model}{options.url}"
        # ❌ model is still in options.json_data when request is sent

Azure's backend then validates the body's model field against actual model names (like gpt-image-1.5), not deployment names (like gpt-image-1-5).

Immediate Workaround

You can manually strip the model parameter using extra_body manipulation or a custom HTTP client:

import httpx
from openai import AzureOpenAI

class AzureModelStripClient(httpx.Client):
    def send(self, request, **kwargs):
        # Strip model from body for deployment-based endpoints
        if b"\"model\":" in request.content and "/deployments/" in str(request.url):
            import json
            body = json.loads(request.content)
            body.pop("model", None)
            request = httpx.Request(
                method=request.method,
                url=request.url,
                headers=request.headers,
                content=json.dumps(body).encode()
            )
        return super().send(request, **kwargs)

client = AzureOpenAI(
    azure_endpoint="https://<resource>.openai.azure.com/",
    api_key="<key>",
    api_version="2024-10-21",
    http_client=AzureModelStripClient()
)

Suggested SDK Fix

In _build_request, after extracting model for URL routing, it should be removed from the body:

def _build_request(self, options, *, retries_taken=0):
    if options.url in _deployments_endpoints and is_mapping(options.json_data):
        model = options.json_data.pop("model", None)  # Changed: pop instead of get
        if model is not None and "/deployments" not in str(self.base_url.path):
            options.url = f"/deployments/{model}{options.url}"
    return super()._build_request(options, retries_taken=retries_taken)

This affects all Azure endpoints where deployment name ≠ model name, not just gpt-image-1.5.

Happy to submit a PR if maintainers agree with the approach.

---

Revolution AI - AI Implementation Specialists