diff --git a/py/src/braintrust/integrations/openai/test_openai.py b/py/src/braintrust/integrations/openai/test_openai.py
index d54b48d36..df54b6b19 100644
--- a/py/src/braintrust/integrations/openai/test_openai.py
+++ b/py/src/braintrust/integrations/openai/test_openai.py
@@ -6,6 +6,7 @@
 import tempfile
 import time
 import zlib
+from types import SimpleNamespace
 
 import openai
 import pytest
@@ -14,6 +15,7 @@
 from braintrust.integrations.openai.tracing import (
     RAW_RESPONSE_HEADER,
     ChatCompletionWrapper,
+    ResponseWrapper,
     _materialize_logged_file_input,
     _process_attachments_in_chat_output,
 )
@@ -2994,3 +2996,21 @@ def test_chat_completion_non_stream_audio_is_materialized_as_attachment(self, me
             filename="generated_audio.wav",
         )
         assert not memory_logger.pop()
+
+
+@pytest.mark.parametrize("event_type", ["response.completed", "response.incomplete", "response.failed"])
+def test_responses_stream_terminal_event_provides_output_and_metrics(event_type):
+    output_item = SimpleNamespace(id="msg_1", type="message", role="assistant", status="incomplete", content=[])
+    usage = SimpleNamespace(input_tokens=10, output_tokens=4, total_tokens=14)
+    all_results = [
+        SimpleNamespace(type="response.created", response=SimpleNamespace(id="resp_1", output=[], usage=None)),
+        SimpleNamespace(
+            type=event_type,
+            response=SimpleNamespace(id="resp_1", output=[output_item], usage=usage),
+        ),
+    ]
+
+    result = ResponseWrapper(None, None)._postprocess_streaming_results(all_results)
+
+    assert result["output"] == [output_item]
+    assert result["metrics"] == {"prompt_tokens": 10, "completion_tokens": 4, "tokens": 14}
diff --git a/py/src/braintrust/integrations/openai/tracing.py b/py/src/braintrust/integrations/openai/tracing.py
index 5b2ed9692..233f0e2a3 100644
--- a/py/src/braintrust/integrations/openai/tracing.py
+++ b/py/src/braintrust/integrations/openai/tracing.py
@@ -1470,6 +1470,12 @@ async def gen():
         return traced_stream
 
 
+# A Responses API stream ends with one of these events. Streams that hit
+# `max_output_tokens` or a content filter end with `response.incomplete`, and
+# all three carry the final `response` object with output and usage.
+_TERMINAL_RESPONSE_EVENT_TYPES = frozenset({"response.completed", "response.incomplete", "response.failed"})
+
+
 class ResponseWrapper:
     def __init__(
         self,
@@ -1652,7 +1658,7 @@ def _postprocess_streaming_results(cls, all_results: list[Any]) -> dict[str, Any
             if (
                 not usage
                 and hasattr(result, "type")
-                and result.type == "response.completed"
+                and result.type in _TERMINAL_RESPONSE_EVENT_TYPES
                 and hasattr(result, "response")
             ):
                 response_dict = _try_to_dict(result.response)
@@ -1682,7 +1688,7 @@ def _postprocess_streaming_results(cls, all_results: list[Any]) -> dict[str, Any
                 output.append(item_data)
                 continue
 
-            if result.type == "response.completed":
+            if result.type in _TERMINAL_RESPONSE_EVENT_TYPES:
                 if hasattr(result, "response") and hasattr(result.response, "output"):
                     if response_dict is None:
                         response_dict = _try_to_dict(result.response)