diff --git a/py/src/braintrust/integrations/openai/test_openai.py b/py/src/braintrust/integrations/openai/test_openai.py
index 439e4c02..7b1f502f 100644
--- a/py/src/braintrust/integrations/openai/test_openai.py
+++ b/py/src/braintrust/integrations/openai/test_openai.py
@@ -525,6 +525,10 @@ def test_openai_responses_web_search_tool_spans(memory_logger):
     tool_spans = _find_spans_by_type(spans, SpanTypeAttribute.TOOL)
 
     assert len(llm_spans) == 1
+    if version_satisfies(openai.__version__, ">=3.41.0"):
+        tool_usage = response.tool_usage.model_dump(exclude_none=True)
+        assert tool_usage["web_search"]["num_requests"] == 1
+        assert _try_to_dict(llm_spans[0]["metadata"]["tool_usage"]) == tool_usage
     tool_span = _find_span_by_name(tool_spans, "web_search_call")
     assert tool_span["span_parents"] == [llm_spans[0]["span_id"]]
     assert tool_span["metadata"]["tool_type"] == "web_search_call"
@@ -566,6 +570,10 @@ def test_openai_responses_web_search_tool_spans_stream(memory_logger):
     tool_spans = _find_spans_by_type(spans, SpanTypeAttribute.TOOL)
 
     assert len(llm_spans) == 1
+    if version_satisfies(openai.__version__, ">=3.41.0"):
+        tool_usage = final_response.tool_usage.model_dump(exclude_none=True)
+        assert tool_usage["web_search"]["num_requests"] == 1
+        assert _try_to_dict(llm_spans[0]["metadata"]["tool_usage"]) == tool_usage
     tool_span = _find_span_by_name(tool_spans, "web_search_call")
     assert tool_span["span_parents"] == [llm_spans[0]["span_id"]]
     assert tool_span["metadata"]["tool_type"] == "web_search_call"
diff --git a/py/src/braintrust/integrations/openai/tracing.py b/py/src/braintrust/integrations/openai/tracing.py
index 38c83570..717f7e62 100644
--- a/py/src/braintrust/integrations/openai/tracing.py
+++ b/py/src/braintrust/integrations/openai/tracing.py
@@ -122,6 +122,7 @@ def start_span(*args, **kwargs):
     "object",
     "background",
     "incomplete_details",
+    "tool_usage",
 )
 
 _EMBEDDING_METADATA_PARAMS = (
@@ -1735,6 +1736,10 @@ def _postprocess_streaming_results(cls, all_results: list[Any]) -> dict[str, Any
 
         for result in all_results:
             usage = getattr(result, "usage", None)
+            if getattr(result, "type", None) in ("response.completed", "response.incomplete", "response.failed"):
+                tool_usage = getattr(getattr(result, "response", None), "tool_usage", None)
+                if tool_usage is not None:
+                    metadata["tool_usage"] = tool_usage
             if (
                 not usage
                 and hasattr(result, "type")