What causes "raw_response.text" in BerriAI/litellm?

Inspect the OpenAIError message: it contains raw_response.text and status_code from Azure — that body is the real upstream error, fix what it reports (auth, endpoint, model).

BerriAI/litellm · error · OpenAIError

raw_response.text

Error message

raw_response.text

What it means

Raised by the Azure AI (MAI) image-generation handler when the upstream HTTP response body cannot be parsed as JSON. The raw response is attempted with raw_response.json(); on any parse failure LiteLLM wraps the body text and HTTP status in an OpenAIError so the caller sees what Azure actually returned instead of a bare JSONDecodeError. Typical causes are non-JSON error pages (HTML auth failures, gateway errors) or truncated bodies.

Source

Thrown at litellm/llms/azure_ai/image_generation/mai_transformation.py:209

            )

    def transform_image_generation_response(
        self,
        model: str,
        raw_response: httpx.Response,
        model_response: ImageResponse,
        logging_obj: "LiteLLMLoggingObj",
        request_data: dict,
        optional_params: dict,
        litellm_params: dict,
        encoding: Any,
        api_key: str | None = None,
        json_mode: bool | None = None,
    ) -> ImageResponse:
        try:
            response: Final = raw_response.json()
        except Exception:
            raise OpenAIError(message=raw_response.text, status_code=raw_response.status_code)

        if "usage" in response:
            response["usage"] = self.normalize_mai_image_usage(response.get("usage"))

        logging_obj.post_call(
            input=request_data.get("prompt", ""),
            api_key=api_key,
            additional_args={"complete_input_dict": request_data},
            original_response=response,
        )

        image_response: Final[ImageResponse] = convert_to_model_response_object(
            response_object=response,
            model_response_object=model_response,
            response_type="image_generation",
        )

        width: Final = optional_params.get("width", self.DEFAULT_WIDTH)

View on GitHub (pinned to 6c2dcb801b)

Solutions

Inspect the OpenAIError message: it contains raw_response.text and status_code from Azure — that body is the real upstream error, fix what it reports (auth, endpoint, model).
Verify AZURE_AI_API_KEY / api_key and AZURE_AI_API_BASE point to a valid Azure MAI endpoint with curl -i to confirm it returns JSON.
If a transient 5xx, retry the call or check Azure status; ensure no corporate proxy is injecting HTML error pages.
Confirm the model string is correct for the MAI image-generation route so you don't land on an error page.

Example fix

# before
response = litellm.image_generation(model="azure_ai/mai/foo", prompt="a cat")

# after
import litellm
from litellm.exceptions import OpenAIError
try:
    response = litellm.image_generation(model="azure_ai/mai/foo", prompt="a cat")
except OpenAIError as e:
    # e.message carries Azure's raw body + status; surface it for diagnosis
    logger.error("Azure MAI image call failed: %s", e)

Defensive patterns

Strategy: try-catch

Try / catch

from litellm.exceptions import OpenAIError
try:
    resp = litellm.image_generation(model="azure_ai/mai/<model>", prompt=p)
except OpenAIError as e:
    # e.message contains Azure's raw body + status code — log and surface it
    log.error("azure mai image failed: %s", e)
    raise

Prevention

Verify AZURE_AI_API_KEY and api_base with a direct curl call that returns JSON before wiring into LiteLLM.
Log the wrapped raw_response.text — it names the real upstream problem.
Keep model strings aligned with the MAI image-generation route.

When it happens

Trigger: Calling image generation via the azure_ai/mai provider (e.g. model 'azure_ai/mai/...') and the endpoint returns a non-JSON body: expired/invalid API key returning an HTML or plain-text 401/403, a 5xx from a proxy/gateway, or a malformed gateway timeout page. Any Exception from raw_response.json() (JSONDecodeError, UnicodeDecodeError) hits this branch.

Common situations: Misconfigured AZURE_AI_API_KEY or wrong api_base pointing at a gateway that returns HTML error pages; Azure outages returning plain-text 503s; regional endpoint mismatches; using a model name the endpoint doesn't host.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/ae54ccd0e7090fb0. Report an issue: GitHub.