BerriAI/litellm · error · ValueError

{model} unable to complete request: {raw_response.incomplete

Error message

{model} unable to complete request: {raw_response.incomplete_details.reason}

What it means

After converting the Responses API output to chat choices, zero choices were produced and raw_response.incomplete_details.reason is set. The provider is telling you it stopped early — e.g. max_output_tokens reached or a content filter fired — so the output contained no convertible message items.

Source

Thrown at litellm/completion_extras/litellm_responses_transformation/transformation.py:766

        if len(output_items) == 0:
            recovered_output_items: Final = self._recover_output_items_from_logging(logging_obj)
            if recovered_output_items:
                output_items = cast(Any, recovered_output_items)
                raw_response.output = cast(Any, recovered_output_items)
                verbose_logger.warning(
                    "Recovered empty Responses API output from raw SSE for model=%s",
                    model,
                )

        # Convert response output to choices using the static helper
        choices: Final = self._convert_response_output_to_choices(
            output_items=output_items,
            handle_raw_dict_callback=self._handle_raw_dict_response_item,
        )

        if len(choices) == 0:
            if raw_response.incomplete_details is not None and raw_response.incomplete_details.reason is not None:
                raise ValueError(f"{model} unable to complete request: {raw_response.incomplete_details.reason}")
            else:
                raise ValueError(f"Unknown items in responses API response: {output_items}")

        setattr(model_response, "choices", choices)

        model_response.model = model

        setattr(
            model_response,
            "usage",
            ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage(raw_response.usage),
        )

        # Preserve hidden params from the ResponsesAPIResponse, especially the headers
        # which contain important provider information like x-request-id
        raw_response_hidden_params: Final = getattr(raw_response, "_hidden_params", {})
        if raw_response_hidden_params:
            if not hasattr(model_response, "_hidden_params") or model_response._hidden_params is None:

View on GitHub (pinned to 6c2dcb801b)

Solutions

Raise or remove max_output_tokens so the model has budget to produce message output
Inspect raw_response.incomplete_details.reason — 'max_output_tokens' means raise the cap; filter reasons mean adjust content
For reasoning models, budget for reasoning tokens explicitly

Example fix

# before
resp = litellm.completion(model="o3", messages=msgs, max_tokens=32)

# after — give the model room for reasoning + message output
resp = litellm.completion(model="o3", messages=msgs, max_tokens=4096)

Defensive patterns

Strategy: validation

Validate before calling

def check_token_headroom(max_tokens: int | None, max_output_tokens: int | None) -> bool:
    cap = max_output_tokens or max_tokens
    return cap is None or cap >= 1024  # leave room for reasoning + message

Try / catch

try:
    resp = litellm.completion(model=m, messages=msgs, max_tokens=cap)
except ValueError as e:
    if "unable to complete request" in str(e) and "max_output_tokens" in str(e):
        resp = litellm.completion(model=m, messages=msgs, max_tokens=cap * 4)
    else:
        raise

Prevention

Set max_tokens generously for reasoning models
Check incomplete_details.reason in responses before bridging
Watch finish_reason/usage in callbacks to catch truncation early

When it happens

Trigger: Calling with max_output_tokens set so low the model never emits a message item; safety-system refusal truncation; providers that stop before any output when filters trigger.

Common situations: Aggressive max_tokens limits in agent loops; content-moderation systems; long reasoning models spending the entire token budget on reasoning items.

Related errors

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/69cca969a5db8c00. Report an issue: GitHub.