diff --git a/litellm/proxy/hooks/dynamic_rate_limiter_v3.py b/litellm/proxy/hooks/dynamic_rate_limiter_v3.py
index a074f02f4e8..0339cf4dfea 100644
--- a/litellm/proxy/hooks/dynamic_rate_limiter_v3.py
+++ b/litellm/proxy/hooks/dynamic_rate_limiter_v3.py
@@ -59,6 +59,10 @@ def _get_priority_settings() -> "PriorityReservationSettings":
     return settings
 
 
+def _is_latin1_encodable(value: object) -> bool:
+    return all(ord(char) < 256 for char in str(value))
+
+
 class _PROXY_DynamicRateLimitHandlerV3(CustomLogger):
     """
     Saturation-aware priority-based rate limiter using v3 infrastructure.
@@ -666,7 +670,13 @@ class _PROXY_DynamicRateLimitHandlerV3(CustomLogger):
             if response_has_hidden_params(response):
                 priority: Final = self._get_priority_from_user_api_key_dict(user_api_key_dict=user_api_key_dict)
                 additional_headers: Final = ensure_response_additional_headers(response)
-                additional_headers["x-litellm-priority"] = priority or "default"
+                priority_header: Final = priority or "default"
+                if _is_latin1_encodable(priority_header):
+                    additional_headers["x-litellm-priority"] = priority_header
+                else:
+                    verbose_proxy_logger.debug(
+                        "Skipping x-litellm-priority header: priority %r is not Latin-1 encodable", priority
+                    )
                 additional_headers["x-litellm-rate-limiter-version"] = "v3"
 
             return response
diff --git a/tests/test_litellm/proxy/hooks/test_dynamic_rate_limiter_v3.py b/tests/test_litellm/proxy/hooks/test_dynamic_rate_limiter_v3.py
index 0cd6b4ede9c..527449bbc48 100644
--- a/tests/test_litellm/proxy/hooks/test_dynamic_rate_limiter_v3.py
+++ b/tests/test_litellm/proxy/hooks/test_dynamic_rate_limiter_v3.py
@@ -1918,3 +1918,30 @@ async def test_post_call_success_hook_leaves_raw_provider_dict_untouched():
     )
 
     assert response == {"id": "msg_123", "type": "message", "role": "assistant", "content": []}
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("team_metadata", "expected_priority_header"),
+    [
+        ({"priority": "优先"}, None),
+        ({"priority": "high"}, "high"),
+        ({}, "default"),
+    ],
+)
+async def test_post_call_success_hook_priority_header_is_always_http_encodable(team_metadata, expected_priority_header):
+    from starlette.responses import Response
+
+    handler = DynamicRateLimitHandler(internal_usage_cache=DualCache())
+    response = {"id": "msg_123", "type": "message", "role": "assistant", "content": [], "_hidden_params": {}}
+
+    await handler.async_post_call_success_hook(
+        data={"model": "anthropic-haiku"},
+        user_api_key_dict=UserAPIKeyAuth(team_id="team-1", team_metadata=team_metadata),
+        response=response,
+    )
+
+    additional_headers = response["_hidden_params"]["additional_headers"]
+    http_response = Response(headers={key: str(value) for key, value in additional_headers.items()})
+    assert http_response.headers.get("x-litellm-priority") == expected_priority_header
+    assert http_response.headers["x-litellm-rate-limiter-version"] == "v3"