microsoft/autogen · error · ValueError

function_call is deprecated and is not supported by this mod

Error message

function_call is deprecated and is not supported by this model client.

What it means

Thrown by OpenAIChatCompletionClient while parsing a completion response when the returned choice.message.function_call is not None. OpenAI's legacy function_call field predates the tool_calls API and this client only supports the modern tool-calling protocol, so any legacy function call in a response is rejected rather than silently mis-parsed.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/models/openai/_openai_client.py:743

        if self._resolved_model is not None:
            if self._resolved_model != result.model:
                warnings.warn(
                    f"Resolved model mismatch: {self._resolved_model} != {result.model}. "
                    "Model mapping in autogen_ext.models.openai may be incorrect. "
                    f"Set the model to {result.model} to enhance token/cost estimation and suppress this warning.",
                    stacklevel=2,
                )

        # Limited to a single choice currently.
        choice: Union[ParsedChoice[Any], ParsedChoice[BaseModel], Choice] = result.choices[0]

        # Detect whether it is a function call or not.
        # We don't rely on choice.finish_reason as it is not always accurate, depending on the API used.
        content: Union[str, List[FunctionCall]]
        thought: str | None = None
        if choice.message.function_call is not None:
            raise ValueError("function_call is deprecated and is not supported by this model client.")
        elif choice.message.tool_calls is not None and len(choice.message.tool_calls) > 0:
            if choice.finish_reason != "tool_calls":
                warnings.warn(
                    f"Finish reason mismatch: {choice.finish_reason} != tool_calls "
                    "when tool_calls are present. Finish reason may not be accurate. "
                    "This may be due to the API used that is not returning the correct finish reason.",
                    stacklevel=2,
                )
            if choice.message.content is not None and choice.message.content != "":
                # Put the content in the thought field.
                thought = choice.message.content
            # NOTE: If OAI response type changes, this will need to be updated
            content = []
            for tool_call in choice.message.tool_calls:
                if not isinstance(tool_call.function.arguments, str):
                    warnings.warn(
                        f"Tool call function arguments field is not a string: {tool_call.function.arguments}."
                        "This is unexpected and may due to the API used not returning the correct type. "

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Point the client at an API version/endpoint that returns tool_calls instead of function_call
2. Update the compatible server/proxy (vLLM, LiteLLM, etc.) to a version that emits the modern tool_calls format
3. If using Azure, set a recent api_version in the client config
4. As a last resort for legacy endpoints, avoid passing tools so the model never emits function calls

Example fix

# before
client = AzureOpenAIChatCompletionClient(azure_endpoint=..., api_version="2023-07-01-preview", ...)

# after
client = AzureOpenAIChatCompletionClient(azure_endpoint=..., api_version="2024-06-01", ...)

try:
    result = await client.create(messages, tools=tools)
except ValueError as e:
    if "function_call is deprecated" in str(e):
        raise RuntimeError(
            "endpoint returns legacy function_call; upgrade the gateway/api_version"
        ) from e
    raise

When it happens

Trigger: The chat completion API (or a compatible endpoint) returns a response using the deprecated function_call field instead of tool_calls. Common with OpenAI-compatible proxies/fine-tunes that emulate the old API, or when a fine-tuned model emits legacy function-call format.

Common situations: Using an OpenAI-compatible gateway (vLLM, older LiteLLM, local servers) that maps tool calls to the legacy function_call response shape; older fine-tuned models trained on the legacy function API; Azure deployments pinned to old API versions.

Related errors

• Function calls are not supported in this context
• Please set OPENAI_API_KEY environment variable.
• Please set OPENAI_API_KEY environment variable.
• Parameter name cannot be null
• param.ParameterType cannot be null

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/0fbb0ebd7d7840e8. Report an issue: GitHub.