diff --git a/strix/config/models.py b/strix/config/models.py
index babb643f..c01b1ccb 100644
--- a/strix/config/models.py
+++ b/strix/config/models.py
@@ -561,6 +561,16 @@ DEFAULT_MODEL_RETRY = ModelRetrySettings(
     ),
 )
 
+_RESPONSE_FAILED_EVENT_TYPE = "response.failed"
+_RESPONSE_FAILED_DEFAULT_MESSAGE = "Response failed"
+_RESPONSE_FAILED_STATUS_CODE = 500
+_RESPONSE_ERROR_ATTR = "error"
+_RESPONSE_ERROR_CODE_ATTR = "code"
+_RESPONSE_ERROR_MESSAGE_ATTR = "message"
+_RESPONSE_EVENT_TYPE_ATTR = "type"
+_RESPONSE_EVENT_RESPONSE_ATTR = "response"
+_LITELLM_RESPONSE_FAILED_PATCH_ATTR = "_strix_raises_response_failed"
+
 RECOMMENDED_MODEL_NAMES = (
     "zai/glm-5.3",
     "zai/glm-5.3-flash",
@@ -661,6 +671,7 @@ def _configure_litellm_compatibility() -> None:
 
     _register_litellm_cost_callback()
     _install_openrouter_stream_cost_capture()
+    _patch_litellm_response_failed_streaming_errors()
 
 
 def _install_openrouter_stream_cost_capture() -> None:
@@ -707,6 +718,64 @@ def _install_openrouter_stream_cost_capture() -> None:
     litellm.OpenrouterConfig = _StrixOpenrouterConfig  # type: ignore[misc]
 
 
+def _patch_litellm_response_failed_streaming_errors() -> None:
+    """Make LiteLLM raise on Responses API ``response.failed`` stream events."""
+    try:
+        from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
+    except Exception:  # noqa: BLE001
+        return
+
+    original_process_chunk = getattr(BaseResponsesAPIStreamingIterator, "_process_chunk", None)
+    if not callable(original_process_chunk):
+        return
+    if getattr(original_process_chunk, _LITELLM_RESPONSE_FAILED_PATCH_ATTR, False):
+        return
+
+    def _strix_process_chunk(self: Any, chunk: Any) -> Any:
+        result = original_process_chunk(self, chunk)
+        _raise_litellm_response_failed(result, self)
+        return result
+
+    setattr(_strix_process_chunk, _LITELLM_RESPONSE_FAILED_PATCH_ATTR, True)
+    BaseResponsesAPIStreamingIterator._process_chunk = _strix_process_chunk  # type: ignore[method-assign]
+
+
+def _raise_litellm_response_failed(chunk: Any, iterator: Any) -> None:
+    if _value(chunk, _RESPONSE_EVENT_TYPE_ATTR) != _RESPONSE_FAILED_EVENT_TYPE:
+        return
+
+    from litellm.exceptions import APIError
+
+    response = _value(chunk, _RESPONSE_EVENT_RESPONSE_ATTR)
+    error = _value(response, _RESPONSE_ERROR_ATTR)
+    message = _response_failed_message(error)
+    raise APIError(
+        status_code=_RESPONSE_FAILED_STATUS_CODE,
+        message=message,
+        llm_provider=str(getattr(iterator, "custom_llm_provider", "") or ""),
+        model=str(getattr(iterator, "model", "") or ""),
+    )
+
+
+def _response_failed_message(error: Any) -> str:
+    message = _value(error, _RESPONSE_ERROR_MESSAGE_ATTR)
+    if not isinstance(message, str) or not message:
+        message = str(error) if error else _RESPONSE_FAILED_DEFAULT_MESSAGE
+
+    code = _value(error, _RESPONSE_ERROR_CODE_ATTR)
+    if isinstance(code, str) and code and code not in message:
+        return f"{message} ({code})"
+    return message
+
+
+def _value(source: Any, key: str) -> Any:
+    if source is None:
+        return None
+    if isinstance(source, dict):
+        return source.get(key)
+    return getattr(source, key, None)
+
+
 OPENROUTER_ATTRIBUTION_HEADERS = {
     "HTTP-Referer": "https://strix.ai",
     "X-Title": "Strix",
diff --git a/tests/test_litellm_response_failed.py b/tests/test_litellm_response_failed.py
new file mode 100644
index 00000000..9b810381
--- /dev/null
+++ b/tests/test_litellm_response_failed.py
@@ -0,0 +1,56 @@
+"""Tests for LiteLLM Responses API failed stream events."""
+
+from __future__ import annotations
+
+import json
+from types import SimpleNamespace
+from typing import Any
+
+import pytest
+from litellm.exceptions import APIError
+from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
+
+from strix.config.models import _configure_litellm_compatibility
+
+
+FAILED_EVENT_TYPE = "response.failed"
+FAILED_MESSAGE = "Insufficient credits"
+FAILED_CODE = "billing_hard_limit_reached"
+FAILED_MODEL = "openai/gpt-5"
+FAILED_PROVIDER = "openai"
+
+
+class _ProviderConfig:
+    def transform_streaming_response(self, **_kwargs: object) -> SimpleNamespace:
+        return SimpleNamespace(
+            type=FAILED_EVENT_TYPE,
+            response=SimpleNamespace(
+                error=SimpleNamespace(message=FAILED_MESSAGE, code=FAILED_CODE)
+            ),
+        )
+
+
+def _make_iterator() -> Any:
+    iterator: Any = BaseResponsesAPIStreamingIterator.__new__(BaseResponsesAPIStreamingIterator)
+    iterator.responses_api_provider_config = _ProviderConfig()
+    iterator.logging_obj = SimpleNamespace()
+    iterator.model = FAILED_MODEL
+    iterator.custom_llm_provider = FAILED_PROVIDER
+    iterator.litellm_metadata = None
+    iterator.completed_response = None
+    iterator._handle_logging_failed_response = lambda: None
+    iterator._handle_logging_completed_response = lambda: None
+    return iterator
+
+
+def test_response_failed_stream_event_is_raised() -> None:
+    _configure_litellm_compatibility()
+    iterator = _make_iterator()
+    payload = json.dumps({"type": FAILED_EVENT_TYPE, "response": {"error": {}}})
+
+    with pytest.raises(APIError) as exc_info:
+        iterator._process_chunk(payload)
+
+    error_text = str(exc_info.value)
+    assert FAILED_MESSAGE in error_text
+    assert FAILED_CODE in error_text