diff --git a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
index eb62b896784..91327e6409e 100644
--- a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
+++ b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
@@ -891,7 +891,9 @@ class HeadroomGuardrail(CustomGuardrail):
         if not has_headroom_retrieve_tool(effective.get("tools")):
             return base_result
         return {  # mutable-ok: the hook contract is a plain dict the router merges into the request kwargs
-            **effective,
+            **{  # mutable-ok: removes stream-only options
+                key: value for key, value in effective.items() if key != "stream_options"
+            },
             "stream": False,
             HEADROOM_CONVERTED_STREAM_KEY: True,
         }
diff --git a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
index bcecb5b27db..406b5df6c00 100644
--- a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
+++ b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
@@ -2155,7 +2155,7 @@ async def test_pre_call_deployment_hook_converts_stream_only_for_ccr_chat_comple
     tools: Optional[list],
     expect_conversion: bool,
 ):
-    kwargs = {"model": "gpt-4o", "stream": stream, "tools": tools}
+    kwargs = {"model": "gpt-4o", "stream": stream, "stream_options": {"include_usage": True}, "tools": tools}
 
     result = await guardrail.async_pre_call_deployment_hook(kwargs=kwargs, call_type=call_type)
 
@@ -2163,12 +2163,15 @@ async def test_pre_call_deployment_hook_converts_stream_only_for_ccr_chat_comple
         assert result is kwargs
         assert HEADROOM_CONVERTED_STREAM_KEY not in kwargs
         assert kwargs["stream"] is stream
+        assert kwargs["stream_options"] == {"include_usage": True}
         return
 
     assert result is not None
     assert result["stream"] is False
+    assert "stream_options" not in result
     assert result[HEADROOM_CONVERTED_STREAM_KEY] is True
     assert kwargs["stream"] is True
+    assert kwargs["stream_options"] == {"include_usage": True}
 
 
 @pytest.mark.asyncio
@@ -2223,6 +2226,7 @@ async def test_pre_call_deployment_hook_converts_stream_after_deployment_level_c
         "model": "gpt-4o",
         "messages": [dict(m) for m in ORIGINAL_MESSAGES],
         "stream": True,
+        "stream_options": {"include_usage": True},
         "guardrails": ["headroom"],
         "metadata": {},
     }
@@ -2238,6 +2242,7 @@ async def test_pre_call_deployment_hook_converts_stream_after_deployment_level_c
     assert result is not None
     assert has_headroom_retrieve_tool(result["tools"])
     assert result["stream"] is False
+    assert "stream_options" not in result
     assert result[HEADROOM_CONVERTED_STREAM_KEY] is True