From bd89d03d2d0c63301186b29bb5ea68f8c2c2bff1 Mon Sep 17 00:00:00 2001
From: hzt <3061613175@qq.com>
Date: Fri, 20 Feb 2026 13:22:58 +0800
Subject: [PATCH] fix(proxy): use completion_model as fallback instead of
 override

completion_model in general_settings always overrode the client-specified
model because it was evaluated first in the Python `or` chain. Reorder
the resolution so the client request model takes priority, matching the
documented behavior: "The default model to use for completions when
model is not specified in the request."

Fixes #21554
---
 litellm/proxy/common_request_processing.py                    | 4 ++--
 .../proxy/pass_through_endpoints/pass_through_endpoints.py    | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py
index 12f7fe2c3cf..8852d9555c7 100644
--- a/litellm/proxy/common_request_processing.py
+++ b/litellm/proxy/common_request_processing.py
@@ -597,10 +597,10 @@ class ProxyBaseLLMRequestProcessing:
             ] = queue_time_seconds
 
         self.data["model"] = (
-            general_settings.get("completion_model", None)  # server default
+            self.data.get("model", None)  # model passed in http request
             or user_model  # model name passed via cli args
             or model  # for azure deployments
-            or self.data.get("model", None)  # default passed in http request
+            or general_settings.get("completion_model", None)  # server default (fallback)
         )
 
         # override with user settings, these are params passed via cli
diff --git a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py
index d1c4b8b3403..64bb74025fe 100644
--- a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py
+++ b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py
@@ -168,9 +168,9 @@ async def chat_completion_pass_through_endpoint(  # noqa: PLR0915
             "Request received by LiteLLM:\n{}".format(json.dumps(data, indent=4)),
         )
         data["model"] = (
-            general_settings.get("completion_model", None)  # server default
+            data.get("model", None)  # model passed in http request
             or user_model  # model name passed via cli args
-            or data.get("model", None)  # default passed in http request
+            or general_settings.get("completion_model", None)  # server default (fallback)
         )
         if user_model:
             data["model"] = user_model