BerriAI/litellm · error · AzureOpenAIError

azure_client is not an instance of AzureOpenAI

Error message

azure_client is not an instance of AzureOpenAI

What it means

After get_azure_openai_client() builds the client for the sync text-completion path, LiteLLM asserts it is an instance of the OpenAI SDK's AzureOpenAI class; otherwise it raises AzureOpenAIError 500. The check catches a misconfigured client factory — e.g. the async class, a plain OpenAI client, or a mock — before calling .completions.with_raw_response.create.

Source

Thrown at litellm/llms/azure/completion/handler.py:154

                        "api_base": api_base,
                        "complete_input_dict": data,
                    },
                )
                if not isinstance(max_retries, int):
                    raise AzureOpenAIError(status_code=422, message="max retries must be an int")
                # init AzureOpenAI Client
                azure_client: Final = self.get_azure_openai_client(
                    api_key=api_key,
                    api_base=api_base,
                    api_version=api_version,
                    client=client,
                    litellm_params=litellm_params,
                    _is_async=False,
                    model=model,
                )

                if not isinstance(azure_client, AzureOpenAI):
                    raise AzureOpenAIError(
                        status_code=500,
                        message="azure_client is not an instance of AzureOpenAI",
                    )

                raw_response: Final = azure_client.completions.with_raw_response.create(**data, timeout=timeout)
                response: Final = raw_response.parse()
                stringified_response: Final = response.model_dump()
                ## LOGGING
                logging_obj.post_call(
                    input=prompt,
                    api_key=api_key,
                    original_response=stringified_response,
                    additional_args={
                        "headers": headers,
                        "api_version": api_version,
                        "api_base": api_base,
                    },
                )

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Do not pass client= for the sync path; let LiteLLM construct the AzureOpenAI client from api_key/api_base/api_version.
2. If you inject a client, ensure it is openai.types/AsyncAzureOpenAI counterpart per path: AzureOpenAI for completion/streaming, AsyncAzureOpenAI for acompletion/astreaming.
3. Pin compatible openai SDK version matching your litellm version so isinstance checks see the same classes.

Example fix

# before
async_client = AsyncAzureOpenAI(...)
llm.completion(..., client=async_client)  # sync path gets async client

# after
sync_client = AzureOpenAI(api_key=..., azure_endpoint=..., api_version=...)
llm.completion(..., client=sync_client)

from openai import AzureOpenAI

def validate_sync_client(client) -> None:
    if client is not None and not isinstance(client, AzureOpenAI):
        raise ConfigError(f"sync path needs AzureOpenAI, got {type(client).__name__}")

from openai import AzureOpenAI
from typing import TypeGuard

def is_sync_azure_client(c: object) -> TypeGuard[AzureOpenAI]:
    return isinstance(c, AzureOpenAI)

try:
    resp = llm.completion(...)
except AzureOpenAIError as e:
    if "not an instance of AzureOpenAI" in str(e):
        raise ConfigError("Wrong client type injected for sync path") from e
    raise

When it happens

Trigger: Passing a custom client= argument that is an AsyncAzureOpenAI or OpenAI instance into the sync path; get_azure_openai_client returning a plain OpenAI client because api_version/base resolution degenerated; test doubles replacing the client.

Common situations: Sharing one client object between sync and async code paths; DI frameworks injecting the wrong client type; upgrading the openai package where class identity moved between module aliases.

Related errors

• azure_client is not an instance of AsyncAzureOpenAI
• azure_client is not an instance of AzureOpenAI or OpenAI
• Unexpected string response from Azure: {response[:500]}
• AzureOpenAI client is not initialized. Make sure api_key is
• AZURE_SENTINEL_DCR_IMMUTABLE_ID is required. Set it as an en

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/0f20f982aca7ad62. Report an issue: GitHub.