diff --git a/litellm/caching/redis_batch.py b/litellm/caching/redis_batch.py
index b3e6d2eb291..d621d09b439 100644
--- a/litellm/caching/redis_batch.py
+++ b/litellm/caching/redis_batch.py
@@ -16,7 +16,7 @@ import logging
 import time
 import weakref
 from collections.abc import Awaitable, Callable, Generator, Mapping, Sequence
-from contextvars import ContextVar, Token
+from contextvars import ContextVar, Token, copy_context
 from dataclasses import dataclass, field
 from datetime import timedelta
 from enum import Enum, auto
@@ -528,6 +528,7 @@ class RequestRedisBatches:
         "_deadline",
         "_deadline_flush",
         "_native",
+        "_native_context",
         "_native_factory",
         "_native_flush_hooks",
         "_native_misses",
@@ -551,6 +552,7 @@ class RequestRedisBatches:
         self._native_factory: Final = native_factory
         self._native_send_factory: Final = native_send_factory
         self._native: NativeRequestRedis | None = None
+        self._native_context: Final = copy_context()
         self._native_flush_hooks: Final[list[Callable[[], None]]] = []  # mutable-ok: append-only registry
         self._native_misses: frozenset[str] = frozenset()
         self._deadline: asyncio.TimerHandle | None = None
@@ -580,9 +582,12 @@ class RequestRedisBatches:
         return batch
 
     def native(self) -> NativeRequestRedis:
-        """The request's native batches, built on first use so requests that touch no native store pay nothing."""
+        """The request's native batches, built on first use so requests that touch no native store pay nothing.
+
+        They keep a copy of the context they are built in, and the cycle collector cannot see through native
+        objects, so they are built in the context this scope was opened from, which cannot reach the scope."""
         if self._native is None:
-            self._native = self._native_factory(self.post_call_deadline)
+            self._native = self._native_context.run(self._native_factory, self.post_call_deadline)
         return self._native
 
     def native_post_call(self) -> NativeRequestRedis:
diff --git a/tests/unit/caching/test_redis_batch.py b/tests/unit/caching/test_redis_batch.py
index 536d510f6d6..158b9754309 100644
--- a/tests/unit/caching/test_redis_batch.py
+++ b/tests/unit/caching/test_redis_batch.py
@@ -3,9 +3,11 @@
 from __future__ import annotations
 
 import asyncio
+import gc
 import hashlib
 import json
 import uuid
+import weakref
 from collections.abc import Awaitable, Callable, Sequence
 from dataclasses import dataclass, field
 from datetime import timedelta
@@ -19,6 +21,7 @@ from litellm._service_logger import ServiceLogging
 from litellm.caching.redis_batch import (
     MIXED_PIPELINE_TARGET,
     RedisBatch,
+    RequestRedisBatches,
     active_request_redis_batch,
     request_redis_batch_scope,
 )
@@ -418,6 +421,17 @@ def test_request_scope_hands_out_one_batch_per_backend_and_nests() -> None:
     assert active_request_redis_batch(cache_a) is None
 
 
+def test_a_request_scope_that_built_its_native_batches_is_freed_once_it_closes() -> None:
+    def serve_request() -> weakref.ref[RequestRedisBatches]:
+        with request_redis_batch_scope() as batches:
+            batches.native()
+            return weakref.ref(batches)
+
+    closed: Final = serve_request()
+    gc.collect()
+    assert closed() is None
+
+
 @pytest.mark.asyncio
 async def test_a_key_an_mget_read_as_absent_stays_known_missing_until_something_sets_it() -> None:
     cache, client = make()