diff --git a/litellm/scheduler.py b/litellm/scheduler.py
index 5309971eeda..41eb0faace4 100644
--- a/litellm/scheduler.py
+++ b/litellm/scheduler.py
@@ -137,7 +137,10 @@ class Scheduler:
             if response is None or not isinstance(response, list):
                 return []
             elif isinstance(response, list):
-                return response
+                # JSON deserializes tuples as lists. heapq requires
+                # consistent types for comparison, so normalize back
+                # to tuples to match what heapq.heappush produces.
+                return [tuple(item) if isinstance(item, list) else item for item in response]
         return self.queue
 
     async def save_queue(self, queue: list, model_name: str) -> None:
diff --git a/tests/test_litellm/test_scheduler_redis_tuple.py b/tests/test_litellm/test_scheduler_redis_tuple.py
new file mode 100644
index 00000000000..43c3344e510
--- /dev/null
+++ b/tests/test_litellm/test_scheduler_redis_tuple.py
@@ -0,0 +1,55 @@
+"""
+Test that Scheduler.get_queue normalizes JSON-deserialized lists back to tuples.
+
+When Redis is used as the cache backend, JSON serialization converts tuples
+to lists. heapq requires consistent types for comparison, so get_queue must
+normalize items back to tuples.
+
+Regression test for https://github.com/BerriAI/litellm/issues/25157
+"""
+
+import pytest
+
+from litellm.scheduler import FlowItem, Scheduler
+
+
+class FakeRedisCache:
+    """Minimal stub that simulates Redis returning JSON-deserialized data."""
+
+    def __init__(self):
+        self.store: dict = {}
+
+    async def async_get_cache(self, key, **kwargs):
+        return self.store.get(key)
+
+    async def async_set_cache(self, key, value, **kwargs):
+        # Simulate JSON round-trip: tuples become lists
+        import json
+
+        self.store[key] = json.loads(json.dumps(value))
+
+
+@pytest.mark.asyncio
+async def test_scheduler_add_request_after_redis_roundtrip():
+    """
+    After a Redis round-trip, queue items are lists (not tuples).
+    Adding a new request should not raise TypeError from heapq comparison.
+    """
+    fake_redis = FakeRedisCache()
+    scheduler = Scheduler(redis_cache=fake_redis)
+
+    # First request — goes into an empty queue, no comparison needed
+    item1 = FlowItem(priority=0, request_id="req-1", model_name="test-model")
+    await scheduler.add_request(item1)
+
+    # Second request — queue from cache has list items, heappush must compare
+    # Without the fix, this raises:
+    #   TypeError: '<' not supported between instances of 'tuple' and 'list'
+    item2 = FlowItem(priority=1, request_id="req-2", model_name="test-model")
+    await scheduler.add_request(item2)
+
+    # Verify both items are in the queue
+    queue = await scheduler.get_queue(model_name="test-model")
+    request_ids = {item[1] for item in queue}
+    assert "req-1" in request_ids
+    assert "req-2" in request_ids