From b6a7f8dcf64503c2f9775c9459426ae103e0b156 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Wed, 28 Oct 2026 06:04:35 -0700
Subject: [PATCH] fix(caching): read the router's model group before the
 caller's metadata

---
 litellm/caching/caching.py                   |  6 ++++--
 tests/unit/caching/test_unit_test_caching.py | 18 +++++++++---------
 2 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/litellm/caching/caching.py b/litellm/caching/caching.py
index 7173f941a8d..e8df765f83d 100644
--- a/litellm/caching/caching.py
+++ b/litellm/caching/caching.py
@@ -456,8 +456,10 @@ class Cache:
         """
         litellm_params: Final[dict] = kwargs.get("litellm_params", {}) or {}
         metadata_sources: Final[tuple[dict, ...]] = (
-            kwargs.get("litellm_metadata" if "litellm_metadata" in kwargs else "metadata") or {},
-            litellm_params.get("litellm_metadata") or litellm_params.get("metadata") or {},
+            kwargs.get("litellm_metadata") or {},
+            kwargs.get("metadata") or {},
+            litellm_params.get("litellm_metadata") or {},
+            litellm_params.get("metadata") or {},
         )
         model_group: Final[str | None] = next(
             (source["model_group"] for source in metadata_sources if source.get("model_group")), None
diff --git a/tests/unit/caching/test_unit_test_caching.py b/tests/unit/caching/test_unit_test_caching.py
index 06b8a4d9cac..fffce818766 100644
--- a/tests/unit/caching/test_unit_test_caching.py
+++ b/tests/unit/caching/test_unit_test_caching.py
@@ -278,7 +278,7 @@ def test_get_model_param_value_reads_model_group_from_litellm_metadata():
     )
 
 
-def test_get_model_param_value_ignores_a_caller_model_group_outside_the_router_bucket():
+def test_get_model_param_value_prefers_the_router_group_over_a_caller_model_group():
     cache = Cache()
     request = {"model": "openai/gpt-5.6", "input": "search this text"}
 
@@ -288,23 +288,23 @@ def test_get_model_param_value_ignores_a_caller_model_group_outside_the_router_b
         )
         == "caller"
     )
-    assert (
-        cache._get_model_param_value({**request, "metadata": {"model_group": "target"}, "litellm_metadata": {}})
-        == "openai/gpt-5.6"
-    )
     assert (
         cache._get_model_param_value(
             {
                 **request,
-                "metadata": {"model_group": "target"},
-                "litellm_metadata": {"user_api_key": "hashed"},
                 "litellm_params": {
                     "metadata": {"model_group": "target"},
-                    "litellm_metadata": {"user_api_key": "hashed"},
+                    "litellm_metadata": {"model_group": "caller"},
                 },
             }
         )
-        == "openai/gpt-5.6"
+        == "caller"
+    )
+    assert (
+        cache._get_model_param_value(
+            {**request, "metadata": {"model_group": "group-a"}, "litellm_metadata": {"attempted_retries": 0}}
+        )
+        == "group-a"
     )
 
 def test_preset_cache_key():