diff --git a/litellm/_lazy_imports.py b/litellm/_lazy_imports.py
index 3bfeba2e394..b8621b10ef9 100644
--- a/litellm/_lazy_imports.py
+++ b/litellm/_lazy_imports.py
@@ -408,14 +408,17 @@ def _lazy_import_http_handlers(name: str) -> Any:
 
     if name == "module_level_aclient":
         # Create an async HTTP client using the factory function
-        from litellm.llms.custom_httpx.http_handler import get_async_httpx_client
+        from litellm.llms.custom_httpx.http_handler import (
+            MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID,
+            get_async_httpx_client,
+        )
 
         # Get timeout from module config (if set)
         timeout = _globals.get("request_timeout")
         params = {"timeout": timeout, "client_alias": "module level aclient"}
         
         # Create the client instance
-        provider_id = cast(Any, "litellm_module_level_client")
+        provider_id = cast(Any, MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID)
         async_client = get_async_httpx_client(
             llm_provider=provider_id,
             params=params,
diff --git a/litellm/llms/custom_httpx/http_handler.py b/litellm/llms/custom_httpx/http_handler.py
index 3dfef07d426..a9abb7427fe 100644
--- a/litellm/llms/custom_httpx/http_handler.py
+++ b/litellm/llms/custom_httpx/http_handler.py
@@ -69,6 +69,7 @@ headers = get_default_headers()
 
 # https://www.python-httpx.org/advanced/timeouts
 _DEFAULT_TIMEOUT = httpx.Timeout(timeout=5.0, connect=5.0)
+MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID = "litellm_module_level_client"
 
 
 def _prepare_request_data_and_content(
@@ -1210,6 +1211,21 @@ def get_async_httpx_client(
 
     Caches the new client and returns it.
     """
+    if llm_provider == MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID:
+        # module_level_aclient is a process-wide singleton and should not be
+        # evicted from bounded in-memory cache under high traffic.
+        if params is not None:
+            handler_params = {
+                k: v for k, v in params.items() if k != "disable_aiohttp_transport"
+            }
+            handler_params["shared_session"] = shared_session
+            return AsyncHTTPHandler(**handler_params)
+
+        return AsyncHTTPHandler(
+            timeout=httpx.Timeout(timeout=600.0, connect=5.0),
+            shared_session=shared_session,
+        )
+
     _params_key_name = ""
     if params is not None:
         for key, value in params.items():
diff --git a/tests/test_litellm/caching/test_redis_connection_pool.py b/tests/test_litellm/caching/test_redis_connection_pool.py
index b8922846e82..a0efe8d0c4b 100644
--- a/tests/test_litellm/caching/test_redis_connection_pool.py
+++ b/tests/test_litellm/caching/test_redis_connection_pool.py
@@ -10,8 +10,13 @@ from unittest.mock import AsyncMock, MagicMock, patch
 import pytest
 import redis.asyncio as async_redis
 
+import litellm
 from litellm._redis import get_redis_async_client, get_redis_connection_pool
 from litellm.caching.llm_caching_handler import LLMClientCache
+from litellm.llms.custom_httpx.http_handler import (
+    MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID,
+    get_async_httpx_client,
+)
 
 
 def test_url_config_uses_passed_pool():
@@ -165,3 +170,48 @@ async def test_eviction_non_closeable_safe():
 
     # If we got here without exception, the test passes
     assert cache.get_cache(key="int-val") == 42
+
+
+@pytest.mark.asyncio
+async def test_module_level_async_client_not_stored_in_bounded_cache():
+    """
+    module_level_aclient is a process-wide singleton and should not be tied to
+    the bounded in-memory cache used for per-provider ephemeral clients.
+    """
+    module_client = litellm.module_level_aclient
+    try:
+        cache_keys = list(getattr(litellm.in_memory_llm_clients_cache, "cache_dict", {}).keys())
+        assert all(
+            MODULE_LEVEL_ASYNC_HTTP_CLIENT_PROVIDER_ID not in key for key in cache_keys
+        )
+        assert module_client.client.is_closed is False
+    finally:
+        await module_client.close()
+        litellm.__dict__.pop("module_level_aclient", None)
+
+
+@pytest.mark.asyncio
+async def test_module_level_async_client_survives_cache_eviction_pressure():
+    """
+    Repro for RuntimeError("client has been closed") seen under traffic:
+    evicting many cached clients must not close module_level_aclient.
+    """
+    module_client = litellm.module_level_aclient
+    cache = litellm.in_memory_llm_clients_cache
+    original_max_size = cache.max_size_in_memory
+    original_default_ttl = cache.default_ttl
+
+    try:
+        cache.max_size_in_memory = 2
+        cache.default_ttl = 3600
+
+        get_async_httpx_client("eviction_provider_1", params={"timeout": None})
+        get_async_httpx_client("eviction_provider_2", params={"timeout": None})
+        await asyncio.sleep(0.05)
+
+        assert module_client.client.is_closed is False
+    finally:
+        cache.max_size_in_memory = original_max_size
+        cache.default_ttl = original_default_ttl
+        await module_client.close()
+        litellm.__dict__.pop("module_level_aclient", None)