BerriAI/litellm · error · CohereError

raw_response.text

Error message

raw_response.text

What it means

In Cohere v1 chat transformation, transform_response tries raw_response.json() and then accesses the 'text' field for the message content. If either fails — non-JSON body or a JSON body missing 'text' (e.g. an error payload) — it raises CohereError with the raw response text and the real HTTP status code. This covers both malformed bodies and API error payloads that bypassed status checks.

Source

Thrown at litellm/llms/cohere/chat/transformation.py:236

    def transform_response(
        self,
        model: str,
        raw_response: httpx.Response,
        model_response: ModelResponse,
        logging_obj: LiteLLMLoggingObj,
        request_data: dict,
        messages: list[AllMessageValues],
        optional_params: dict,
        litellm_params: dict,
        encoding: Any,
        api_key: str | None = None,
        json_mode: bool | None = None,
    ) -> ModelResponse:
        try:
            raw_response_json: Final = raw_response.json()
            model_response.choices[0].message.content = raw_response_json["text"]
        except Exception:
            raise CohereError(message=raw_response.text, status_code=raw_response.status_code)

        ## ADD CITATIONS
        if "citations" in raw_response_json:
            setattr(model_response, "citations", raw_response_json["citations"])

        ## Tool calling response
        cohere_tools_response: Final = raw_response_json.get("tool_calls", None)
        if cohere_tools_response is not None and cohere_tools_response != []:
            # convert cohere_tools_response to OpenAI response format
            tool_calls: Final = []
            for tool in cohere_tools_response:
                function_name = tool.get("name", "")
                generation_id = tool.get("generation_id", "")
                parameters = tool.get("parameters", {})
                tool_call = {
                    "id": f"call_{generation_id}",
                    "type": "function",
                    "function": {

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Inspect error.message (raw body) and error.status_code: Cohere error payloads state the actual problem (e.g. invalid model, invalid API key).
2. Switch to a current model (command-r-plus / command-a) via the v2 path — the v1 'text' shape is legacy.
3. If api_base was overridden, remove it or point it at the genuine Cohere endpoint.
4. Retry once for transient truncation; if persistent, enable verbose logging to capture the full body.

Example fix

# before (v1 path, may miss 'text')
resp = litellm.completion(model="cohere/command-nightly", messages=[...])

# after (v2-capable model)
resp = litellm.completion(model="cohere/command-r-plus", messages=[...])

from litellm.exceptions import CohereError

try:
    resp = litellm.completion(model="cohere/command-r-plus", messages=msgs)
except CohereError as e:
    if e.status_code == 200 or "text" not in (e.message or ""):
        # schema/shape issue: prefer v2 models instead of retrying blind
        raise RuntimeError(f"Unexpected Cohere v1 body: {e.message}") from e
    raise

When it happens

Trigger: Calling cohere/* v1 chat models (command, command-light, command-nightly) when the response body lacks a 'text' field: error JSON like {"message": "..."} returned with 200/4xx, HTML from a proxy, or a truncated body.

Common situations: Cohere returning an error object without 'text' (invalid model name, bad parameters); api_base customized to a gateway returning unexpected JSON; deprecated v1 endpoint shape changes. Note Cohere has deprecated v1 in favor of v2 models — using a v2-only model through the v1 path triggers this.

Related errors

• raw_response.text
• Failed to decode JSON from chunk: {chunk}
• Error parsing chunk: {e}, Received chunk: {chunk}
• {service_name} returned non-dict JSON ({type(result).__name_
• CohereException - {original_exception.message}

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/6a9548daef5035f0. Report an issue: GitHub.