microsoft/autogen · error · ValueError

Model does not support JSON output.

Error message

Model does not support JSON output.

What it means

At construction, if the resolved model_info has json_output: False but create_args contains response_format={'type': 'json_object'}, the client raises ValueError — it refuses to request JSON mode from a model declared incapable of it. This validates constructor-time config, before any request is made.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/models/ollama/_ollama_client.py:498

            self._model_info = info
        elif model_capabilities is None and model_info is not None:
            self._model_info = model_info

        self._resolved_model: Optional[str] = None
        self._model_class: Optional[str] = None
        if "model" in create_args:
            self._resolved_model = create_args["model"]
            self._model_class = _model_info.resolve_model_class(create_args["model"])

        if (
            not self._model_info["json_output"]
            and "response_format" in create_args
            and (
                isinstance(create_args["response_format"], dict)
                and create_args["response_format"]["type"] == "json_object"
            )
        ):
            raise ValueError("Model does not support JSON output.")

        self._create_args = create_args
        self._total_usage = RequestUsage(prompt_tokens=0, completion_tokens=0)
        self._actual_usage = RequestUsage(prompt_tokens=0, completion_tokens=0)
        # Ollama doesn't have IDs for tools, so we just increment a counter
        self._tool_id = 0

    @classmethod
    def create_from_config(cls, config: Dict[str, Any]) -> ChatCompletionClient:
        return OllamaChatCompletionClient(**config)

    def get_create_args(self) -> Mapping[str, Any]:
        return self._create_args

    def _process_create_args(
        self,
        messages: Sequence[LLMMessage],
        tools: Sequence[Tool | ToolSchema],

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Remove response_format / json_output for models without JSON support, or switch to a model that supports it (e.g. llama3.1+)
2. Set json_output: True in a custom model_info if you are certain the underlying Ollama model handles JSON
3. Prefer the json_output parameter over raw response_format (response_format triggers deprecation warnings elsewhere in this client)

Example fix

# before
client = OllamaChatCompletionClient(model="llama2", response_format={"type": "json_object"})

# after
client = OllamaChatCompletionClient(model="llama3.1", json_output=True)

if create_args.get("response_format", {}).get("type") == "json_object" and not info["json_output"]:
    del create_args["response_format"]  # or pick a JSON-capable model
client = OllamaChatCompletionClient(client=oc, create_args=create_args, model_info=info)

def wants_json_mode(create_args: Mapping[str, Any]) -> bool:
    rf = create_args.get("response_format")
    return isinstance(rf, dict) and rf.get("type") == "json_object"

try:
    client = OllamaChatCompletionClient(model=name, response_format={"type": "json_object"})
except ValueError as e:
    if "does not support JSON" in str(e):
        client = OllamaChatCompletionClient(model="llama3.1")  # JSON-capable tag
    else:
        raise

When it happens

Trigger: OllamaChatCompletionClient(model='llama2', response_format={'type': 'json_object'}) where the registry marks llama2 as json_output: False; passing a custom model_info with json_output: False together with a JSON response_format.

Common situations: Reusing an OpenAI-style response_format config with a smaller Ollama model; defaulting response_format globally for structured-output pipelines and switching models underneath.

Related errors

• model_info is required when model name is not a valid OpenAI
• response_format and json_output cannot be set to a Pydantic
• json_output and format cannot be set at the same time. Use j
• model_info is required for AzureAIChatCompletionClient
• Model does not support JSON output

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/3f650fa840cb963b. Report an issue: GitHub.