fix(vertex-ai): reuse anthropic messages config instances

Cache provider config lookups for Vertex Anthropic messages so repeated requests reuse the same config object and preserve credential cache state. Add a regression test to catch any future loss of config reuse.

Made-with: Cursor

litellm/utils.py

@@ -8410,6 +8410,17 @@ class ProviderConfigManager:
         model: str,
         provider: LlmProviders,
     ) -> Optional[BaseAnthropicMessagesConfig]:
+        return ProviderConfigManager._get_provider_anthropic_messages_config_cached(
+            model=model, provider=provider
+        )
+
+    @staticmethod
+    @lru_cache(maxsize=DEFAULT_MAX_LRU_CACHE_SIZE)
+    def _get_provider_anthropic_messages_config_cached(
+        model: str,
+        provider: LlmProviders,
+    ) -> Optional[BaseAnthropicMessagesConfig]:
+        model_lower = model.lower()
         if litellm.LlmProviders.ANTHROPIC == provider:
             return litellm.AnthropicMessagesConfig()
         # The 'BEDROCK' provider corresponds to Amazon's implementation of Anthropic Claude v3.
@@ -8419,14 +8430,14 @@ class ProviderConfigManager:
 
             return BedrockModelInfo.get_bedrock_provider_config_for_messages_api(model)
         elif litellm.LlmProviders.VERTEX_AI == provider:
-            if "claude" in model.lower():
+            if "claude" in model_lower:
                 from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.experimental_pass_through.transformation import (
                     VertexAIPartnerModelsAnthropicMessagesConfig,
                 )
 
                 return VertexAIPartnerModelsAnthropicMessagesConfig()
         elif litellm.LlmProviders.AZURE_AI == provider:
-            if "claude" in model.lower():
+            if "claude" in model_lower:
                 from litellm.llms.azure_ai.anthropic.messages_transformation import (
                     AzureAnthropicMessagesConfig,
                 )

tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py

@@ -311,3 +311,29 @@ def test_transform_anthropic_messages_request_removes_scope_from_cache_control()
     # scope removed from message content
     assert "scope" not in result["messages"][0]["content"][0]["cache_control"]
     assert result["messages"][0]["content"][0]["cache_control"]["type"] == "ephemeral"
+
+
+def test_provider_config_manager_reuses_vertex_anthropic_messages_config_instance():
+    """
+    Regression test: repeated provider config lookups for the same Vertex Claude model
+    should return the same config instance (which preserves auth cache state).
+    """
+    import litellm
+    from litellm.utils import ProviderConfigManager
+
+    ProviderConfigManager._get_provider_anthropic_messages_config_cached.cache_clear()
+    try:
+        first_config = ProviderConfigManager.get_provider_anthropic_messages_config(
+            model="claude-opus-4-6",
+            provider=litellm.LlmProviders.VERTEX_AI,
+        )
+        second_config = ProviderConfigManager.get_provider_anthropic_messages_config(
+            model="claude-opus-4-6",
+            provider=litellm.LlmProviders.VERTEX_AI,
+        )
+
+        assert isinstance(first_config, VertexAIPartnerModelsAnthropicMessagesConfig)
+        assert isinstance(second_config, VertexAIPartnerModelsAnthropicMessagesConfig)
+        assert first_config is second_config
+    finally:
+        ProviderConfigManager._get_provider_anthropic_messages_config_cached.cache_clear()