diff --git a/litellm/caching/caching_handler.py b/litellm/caching/caching_handler.py
index 1d80d6952a4..5738a9c2813 100644
--- a/litellm/caching/caching_handler.py
+++ b/litellm/caching/caching_handler.py
@@ -182,14 +182,16 @@ def _should_defer_streaming_cache_hit_callbacks(*, cached_result: object) -> boo
     spend and callback records. A plain (non-stream) replay logs here, since nothing
     else will.
     """
-    from litellm.llms.anthropic.messages.native.response_cache import (
-        CachedAnthropicMessagesStreamIterator,
-    )
     from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
 
-    return isinstance(
-        cached_result, (CustomStreamWrapper, BaseResponsesAPIStreamingIterator, CachedAnthropicMessagesStreamIterator)
-    )
+    streaming_types: Final[tuple[type, ...]] = (CustomStreamWrapper, BaseResponsesAPIStreamingIterator)
+    try:
+        from litellm.llms.anthropic.messages.native.response_cache import (
+            CachedAnthropicMessagesStreamIterator,
+        )
+    except ImportError:
+        return isinstance(cached_result, streaming_types)
+    return isinstance(cached_result, (*streaming_types, CachedAnthropicMessagesStreamIterator))
 
 
 def _prompt_tokens_details_as_mapping(details: "PromptTokensDetailsWrapper") -> Mapping[str, object]:
diff --git a/tests/unit/caching/test_caching_handler.py b/tests/unit/caching/test_caching_handler.py
index 51754c80dd6..145a6e78879 100644
--- a/tests/unit/caching/test_caching_handler.py
+++ b/tests/unit/caching/test_caching_handler.py
@@ -1,4 +1,5 @@
 import time
+import sys
 import traceback
 from litellm._uuid import uuid
 
@@ -1281,6 +1282,17 @@ def test_should_defer_streaming_cache_hit_callbacks_for_any_streaming_request():
     assert _should_defer_streaming_cache_hit_callbacks(cached_result={"id": "msg_1"}) is False
 
 
+def test_should_defer_streaming_cache_hit_callbacks_without_proxy_extra(monkeypatch):
+    monkeypatch.setitem(sys.modules, "litellm.llms.anthropic.messages.native.response_cache", None)
+    logging_obj = MagicMock()
+    logging_obj.model_call_details = {}
+    stream_replay = CustomStreamWrapper(
+        completion_stream=iter(()), model="gpt-4o", logging_obj=logging_obj
+    )
+    assert _should_defer_streaming_cache_hit_callbacks(cached_result=stream_replay) is True
+    assert _should_defer_streaming_cache_hit_callbacks(cached_result=ModelResponse()) is False
+
+
 @pytest.mark.asyncio
 async def test_async_get_cache_defers_streaming_completion_hit_callbacks():
     litellm.set_verbose = True