diff --git a/litellm/completion_extras/litellm_responses_transformation/transformation.py b/litellm/completion_extras/litellm_responses_transformation/transformation.py
index 85fb0bc8dc6..99a600ce78a 100644
--- a/litellm/completion_extras/litellm_responses_transformation/transformation.py
+++ b/litellm/completion_extras/litellm_responses_transformation/transformation.py
@@ -1546,6 +1546,18 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator):
                 ],
                 usage=usage,
             )
+        elif event_type in ("response.failed", "error"):
+            failure_response_data = parsed_chunk.get("response") or {}
+            if not isinstance(failure_response_data, dict):
+                failure_response_data = {}
+            error_data = failure_response_data.get("error") or parsed_chunk.get("error") or parsed_chunk
+            error_message = error_data.get("message") if isinstance(error_data, dict) else None
+            raise litellm.APIError(
+                status_code=500,
+                message=error_message or "Responses API stream failed",
+                llm_provider="",
+                model=failure_response_data.get("model", ""),
+            )
         else:
             pass
         # For any unhandled event types, create a minimal valid chunk or skip
diff --git a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
index 21b60d7a216..b03ba421269 100644
--- a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
+++ b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
@@ -21,6 +21,37 @@ if TYPE_CHECKING:
     from litellm.types.utils import ModelResponse
 
 
+@pytest.mark.parametrize(
+    ("chunk", "message"),
+    [
+        (
+            {
+                "type": "response.failed",
+                "response": {
+                    "model": "gpt-5.6-luna",
+                    "error": {"code": "server_error", "message": "upstream failed"},
+                },
+            },
+            "upstream failed",
+        ),
+        ({"type": "error", "code": "server_error", "message": "stream failed"}, "stream failed"),
+        (
+            {"type": "response.failed", "response": "invalid", "error": {"message": "malformed failure"}},
+            "malformed failure",
+        ),
+    ],
+)
+def test_response_stream_failure_raises_api_error(chunk, message):
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        OpenAiResponsesToChatCompletionStreamIterator,
+    )
+
+    with pytest.raises(litellm.APIError, match=message) as exc_info:
+        OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(chunk)
+
+    assert exc_info.value.status_code == 500
+
+
 def test_convert_chat_completion_messages_to_responses_api_image_input():
     from litellm.completion_extras.litellm_responses_transformation.transformation import (
         LiteLLMResponsesTransformationHandler,