### What happened?

When calling `/chat/completions` with a model backed by the Responses API, LiteLLM converts assistant history messages into Responses `input` items. The generated `output_text` content blocks do not include the `annotations` field:

```python # litellm/completion_extras/litellm_responses_transformation/transformation.py def _convert_content_str_to_input_text(self, content: str, role: str) -> Dict[str, Any]: if role == "user" or role == "system" or role == "tool": return {"type": "input_text", "text": content} else: return {"type": "output_text", "text": content} # <-- no "annotations" ```

OpenAI's public API tolerates the missing field, but stricter Responses-compatible backends (e.g. Azure-style validators) reject the payload, since the `output_text` schema marks `annotations` as required:

``` litellm.BadRequestError: OpenAIException - { "error": { "code": "invalid_payload", "message": "Required property 'annotations' is missing", "param": "input[1].content[0].annotations", "type": "invalid_request_error", ... ```

The first turn works (no assistant history); every subsequent turn fails once an assistant message …