BerriAI/litellm · error · AnthropicError

{completion_response["error"]}

Error message

{completion_response["error"]}

What it means

After a non-streaming call, LiteLLM checks the parsed response body for an 'error' key; if present it raises AnthropicError with that error payload as the message and the HTTP status code from the raw response. This is the Anthropic API's own error (overloaded, invalid_request, authentication, rate limit) surfaced as a Python exception — check status_code to classify it.

Source

Thrown at litellm/llms/anthropic/chat/transformation.py:2352

            provider_specific_fields["compaction_blocks"] = compaction_blocks

        return provider_specific_fields

    def transform_parsed_response(
        self,
        completion_response: dict,
        raw_response: httpx.Response,
        model_response: ModelResponse,
        json_mode: bool | None = None,
        prefix_prompt: str | None = None,
        speed: str | None = None,
        tool_name_reverse_map: dict[str, str] | None = None,
    ):
        _hidden_params: Final[dict] = {}
        _hidden_params["additional_headers"] = process_anthropic_headers(dict(raw_response.headers))
        if "error" in completion_response:
            response_headers: Final = getattr(raw_response, "headers", None)
            raise AnthropicError(
                message=str(completion_response["error"]),
                status_code=raw_response.status_code,
                headers=response_headers,
            )

        (
            text_content,
            citations,
            thinking_blocks,
            reasoning_content,
            tool_calls,
            web_search_results,
            tool_results,
            compaction_blocks,
        ) = self.extract_response_content(completion_response=completion_response)

        # Reverse-map rewritten tool names back to caller's originals so a
        # downstream OpenAI-style dispatcher can match on the registered name.

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Inspect exception.status_code and the message body: 401 -> fix ANTHROPIC_API_KEY; 429 -> back off / raise limits; 529 -> retry with jitter; 400 -> fix request params.
2. For 429/529 use litellm's built-in retries (num_retries) or exponential backoff around the call.
3. Verify key and base URL env vars are set and current.
4. Reduce max_tokens or trim input if the error mentions context length.

Example fix

# before
resp = litellm.completion(model="anthropic/claude-sonnet-4-5", messages=msgs)

# after
import litellm, time
for attempt in range(5):
    try:
        resp = litellm.completion(
            model="anthropic/claude-sonnet-4-5", messages=msgs,
            num_retries=3,  # handles 429/529 automatically
        )
        break
    except litellm.exceptions.AnthropicError as e:
        if e.status_code in (429, 529) and attempt < 4:
            time.sleep(2 ** attempt)
            continue
        raise

from litellm.exceptions import AnthropicError
try:
    resp = litellm.completion(model=MODEL, messages=messages, num_retries=3)
except AnthropicError as e:
    if e.status_code == 401:
        rotate_key_and_alert()
    elif e.status_code in (429, 529):
        schedule_retry_with_backoff()
    elif e.status_code == 400:
        fix_request_from_message(e.message)
    raise

When it happens

Trigger: Any non-streaming anthropic/ completion where the API responds with an error JSON: 401 bad API key, 429 rate limit, 529 overloaded, 400 invalid_request_error (bad params, context length), or a proxy returning Anthropic-shaped error bodies.

Common situations: Hitting rate limits under load; expired/rotated API keys; requests exceeding max tokens/context; Anthropic capacity events (529); misconfigured LiteLLM proxy keys.

Related errors

• Missing Anthropic API Key
• Anthropic API key is required. Set ANTHROPIC_API_KEY or ANTH
• ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN is required for Sk
• BFL initial request failed: {initial_response.text}
• Failed to transform Braintrust response: {str(e)}

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/1db1ea02a747e960. Report an issue: GitHub.