Fix: base_model name for body and deplyment name in URL

[Sameerlite]

Relevant issues

Fixes #20741

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

model_list:
  - model_name: gpt-image-1.5 
    litellm_params:
      base_model: gpt-image-1.5 
      model: azure/gpt-image-15 # deployment name is different 
      api_base: https://openai-gpt-image-1-5-test-v-1.openai.azure.com/

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 9, 2026 11:45am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR updates Azure image generation so the request body can use litellm_params.base_model while the request URL is constructed with the Azure deployment name (model). It also adds unit tests intended to verify that split.

The changes fit into the existing Azure image generation flow in AzureChatCompletion by (1) overriding the JSON body’s model field when base_model is configured and (2) threading the deployment name into create_azure_base_url() for both sync and async paths.

Confidence Score: 2/5

• This PR should not be merged until the new null-deref in Azure image generation and the ineffective tests are fixed.
• The core behavior change is reasonable, but image_generation() can now crash when litellm_params is None, and the added tests include a non-async async-marked test plus exception swallowing that can let regressions slip through.
• litellm/llms/azure/azure.py; tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py

Important Files Changed

Filename	Overview
litellm/llms/azure/azure.py	Adjusts Azure image generation to use deployment name for URL and base_model for request body, but introduces a null-deref crash when litellm_params is None (litellm_params.get(...) in image_generation).
tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py	Adds sync/async tests for base_model vs deployment URL behavior, but one async test is defined as non-async and both tests swallow exceptions, risking false positives.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant AzureChatCompletion as AzureChatCompletion
    participant AzureBaseURL as create_azure_base_url
    participant SyncReq as make_sync_azure_httpx_request
    participant AsyncReq as make_async_azure_httpx_request

    alt Sync image_generation()
        Caller->>AzureChatCompletion: image_generation(prompt, model=deployment, litellm_params.base_model)
        AzureChatCompletion->>AzureChatCompletion: extra_body flatten + data.model = base_model || deployment
        AzureChatCompletion->>AzureBaseURL: create_azure_base_url(model=deployment)
        AzureBaseURL-->>AzureChatCompletion: img_gen_api_base (/openai/deployments/{deployment}/images/generations)
        AzureChatCompletion->>SyncReq: POST(api_base=img_gen_api_base, json=data)
        SyncReq-->>AzureChatCompletion: httpx.Response
        AzureChatCompletion-->>Caller: ImageResponse
    end

    alt Async aimage_generation()
        Caller->>AzureChatCompletion: aimage_generation(data.model=base_model, model=deployment)
        AzureChatCompletion->>AzureBaseURL: create_azure_base_url(model=deployment)
        AzureBaseURL-->>AzureChatCompletion: img_gen_api_base
        AzureChatCompletion->>AsyncReq: POST(api_base=img_gen_api_base, json=data)
        AsyncReq-->>AzureChatCompletion: httpx.Response
        AzureChatCompletion-->>Caller: ImageResponse
    end

Loading

[greptile-apps]

_{2 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Async test never runs

The async test is marked with @pytest.mark.asyncio but the function is defined with def (not async def). Pytest will treat it as a normal test and won’t await anything, so this doesn’t actually validate the async path.

This should be async def test_azure_aimage_generation_base_model_vs_deployment_name():.

[Sameerlite]

It is async def test_azure_aimage_generation_base_model_vs_deployment_name

[greptile-apps]

Test can falsely pass

Both new tests wrap the call in try/except Exception: pass and then only assert mock_request.called. This can still pass even if image_generation() fails before constructing the URL/body you want to verify (e.g., raises due to unrelated bug), and can hide regressions.

Remove the blanket exception handler (or assert on the specific expected exception) so the test actually fails when the call path breaks.

[greptile-apps]

Additional Comments (1)

litellm/llms/azure/azure.py
Crash when litellm_params None

image_generation() now calls litellm_params.get(...) unconditionally. Since litellm_params is typed/used as optional and can be None at runtime, this will raise AttributeError: 'NoneType' object has no attribute 'get' on image generation calls that don’t pass litellm_params.

This needs a litellm_params = litellm_params or {} (or equivalent guard) before accessing .get().

[ceng-p]

Looks like a decent work around, thanks