use generic method to remove the unsupported param

litellm/llms/anthropic/chat/transformation.py

@@ -736,15 +736,11 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
         ):
             optional_params["metadata"] = {"user_id": _litellm_metadata["user_id"]}
 
-        if "prompt_cache_key" in optional_params:
-            optional_params.pop("prompt_cache_key")
-
         data = {
             "model": model,
             "messages": anthropic_messages,
             **optional_params,
         }
-
         return data
 
     def _transform_response_for_json_mode(

litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py

@@ -80,7 +80,6 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig):
 
         _anthropic_request.pop("model", None)
         _anthropic_request.pop("stream", None)
-        _anthropic_request.pop("prompt_cache_key", None) 
         if "anthropic_version" not in _anthropic_request:
             _anthropic_request["anthropic_version"] = self.anthropic_version
 

litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py

@@ -129,11 +129,7 @@ class AmazonAnthropicClaudeMessagesConfig(
         if "model" in anthropic_messages_request:
             anthropic_messages_request.pop("model", None)
 
-        # 4. `prompt_cache_key` is not allowed in request body for bedrock invoke
-        if "prompt_cache_key" in anthropic_messages_request:
-            anthropic_messages_request.pop("prompt_cache_key", None)
-            
-        # 5. Handle anthropic_beta from user headers
+        # 4. Handle anthropic_beta from user headers
         anthropic_beta_list = get_anthropic_beta_from_headers(headers)
         if anthropic_beta_list:
             anthropic_messages_request["anthropic_beta"] = anthropic_beta_list

litellm/llms/openai/chat/gpt_transformation.py

@@ -160,6 +160,7 @@ class OpenAIGPTConfig(BaseLLMModelInfo, BaseConfig):
             "web_search_options",
             "service_tier",
             "safety_identifier",
+            "prompt_cache_key",
         ]  # works across all models
 
         model_specific_params = []

litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py

@@ -68,7 +68,6 @@ class VertexAIAnthropicConfig(AnthropicConfig):
         )
 
         data.pop("model", None)  # vertex anthropic doesn't accept 'model' parameter
-        data.pop("prompt_cache_key", None)  # vertex anthropic doesn't accept 'prompt_cache_key' parameter
         return data
 
     def transform_response(

litellm/responses/litellm_completion_transformation/transformation.py

@@ -12,6 +12,9 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
 from litellm.responses.litellm_completion_transformation.session_handler import (
     ResponsesSessionHandler,
 )
+from litellm.litellm_core_utils.get_supported_openai_params import (
+    get_supported_openai_params,
+)
 from litellm.types.llms.openai import (
     AllMessageValues,
     ChatCompletionImageObject,
@@ -94,6 +97,23 @@ class LiteLLMCompletionResponsesConfig:
             "user",
         ]
 
+    @staticmethod
+    def filter_unsupported_params(
+        params: dict,
+        model: str,
+        custom_llm_provider: Optional[str] = None,
+    ) -> None:
+        """Remove params not supported by the provider."""
+        supported_params = get_supported_openai_params(
+            model=model, custom_llm_provider=custom_llm_provider
+        )
+        if supported_params:
+            keys_to_remove = [
+                key for key in params.keys() if key not in supported_params
+            ]
+            for key in keys_to_remove:
+                params.pop(key, None)
+
     @staticmethod
     def transform_responses_api_request_to_chat_completion_request(
         model: str,
@@ -118,6 +138,12 @@ class LiteLLMCompletionResponsesConfig:
                 text_param
             )
 
+        LiteLLMCompletionResponsesConfig.filter_unsupported_params(
+            params=responses_api_request,
+            model=model,
+            custom_llm_provider=custom_llm_provider,
+        )
+
         litellm_completion_request: dict = {
             "messages": LiteLLMCompletionResponsesConfig.transform_responses_api_input_to_messages(
                 input=input,
@@ -136,6 +162,7 @@ class LiteLLMCompletionResponsesConfig:
             "service_tier": kwargs.get("service_tier"),
             "web_search_options": web_search_options,
             "response_format": response_format,
+            "prompt_cache_key": responses_api_request.get("prompt_cache_key"),
             # litellm specific params
             "custom_llm_provider": custom_llm_provider,
             "extra_headers": extra_headers,
@@ -157,7 +184,7 @@ class LiteLLMCompletionResponsesConfig:
         litellm_completion_request = {
             k: v for k, v in litellm_completion_request.items() if v is not None
         }
-
+        print(f"litellm_completion_request: {litellm_completion_request.keys()}")
         return litellm_completion_request
 
     @staticmethod

litellm/responses/main.py

@@ -134,6 +134,7 @@ async def aresponses_api_with_mcp(
     top_p: Optional[float] = None,
     truncation: Optional[Literal["auto", "disabled"]] = None,
     user: Optional[str] = None,
+    prompt_cache_key: Optional[str] = None,
     # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
     # The extra values given here take precedence over values defined on the client or passed to this method.
     extra_headers: Optional[Dict[str, Any]] = None,
@@ -489,6 +490,7 @@ def responses(
     user: Optional[str] = None,
     service_tier: Optional[str] = None,
     safety_identifier: Optional[str] = None,
+    prompt_cache_key: Optional[str] = None,
     # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
     # The extra values given here take precedence over values defined on the client or passed to this method.
     extra_headers: Optional[Dict[str, Any]] = None,
@@ -571,6 +573,7 @@ def responses(
                 top_p=top_p,
                 truncation=truncation,
                 user=user,
+                prompt_cache_key=prompt_cache_key,
                 extra_headers=extra_headers,
                 extra_query=extra_query,
                 extra_body=extra_body,

proxy_server_config.yaml

@@ -94,9 +94,13 @@ model_list:
       rpm: 1000
     model_info:
       health_check_timeout: 1
-  - model_name: "*"
+  - model_name: "gpt-4o-mini"
     litellm_params:
-      model: openai/*
+      model: openai/gpt-4o-mini
+      api_key: os.environ/OPENAI_API_KEY
+  - model_name: claude-sonnet-4
+    litellm_params:
+      model: openai/gpt-4o-mini
       api_key: os.environ/OPENAI_API_KEY
   
 

tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -372,42 +372,18 @@ def test_prompt_cache_key_removed_from_request():
     
     This prevents the error: "prompt_cache_key: Extra inputs are not permitted"
     
-    Related issue: https://github.com/BerriAI/litellm/issues/xxxxx
     """
     config = AnthropicConfig()
     
-    # Simulate a request with prompt_cache_key
-    messages = [
-        {"role": "user", "content": "Hello, how are you?"}
-    ]
-    
-    optional_params = {
-        "max_tokens": 100,
-        "temperature": 0.7,
-        "prompt_cache_key": "test-cache-key-12345",  # This should be removed
-    }
-    
-    litellm_params = {}
-    headers = {"anthropic-version": "2023-06-01"}
-    
-    # Transform the request
-    result = config.transform_request(
-        model="claude-3-7-sonnet-20250219",
-        messages=messages,
-        optional_params=optional_params,
-        litellm_params=litellm_params,
-        headers=headers,
+    # Verify that prompt_cache_key is NOT in Anthropic's supported params
+    supported_params = config.get_supported_openai_params("claude-3-7-sonnet-20250219")
+    assert "prompt_cache_key" not in supported_params, (
+        "prompt_cache_key should not be in Anthropic's supported params list"
     )
     
-    # Verify prompt_cache_key is NOT in the result
-    assert "prompt_cache_key" not in result, "prompt_cache_key should be removed from the request"
+    # Verify other common parameters are present
+    assert "max_tokens" in supported_params
+    assert "temperature" in supported_params
+    assert "tools" in supported_params
     
-    # Verify other parameters are still present
-    assert "max_tokens" in result
-    assert result["max_tokens"] == 100
-    assert "temperature" in result
-    assert result["temperature"] == 0.7
-    assert "model" in result
-    assert "messages" in result
-    
-    print("✅ Test passed: prompt_cache_key successfully removed from Anthropic request")
+    print("✅ Test passed: prompt_cache_key is not in Anthropic's supported params")

tests/test_litellm/llms/bedrock/chat/invoke_transformations/test_bedrock_chat_invoke_transformations_anthropic_claude3_transformation.py

@@ -33,47 +33,15 @@ def test_prompt_cache_key_removed_from_bedrock_request():
     """
     config = AmazonAnthropicClaudeConfig()
     
-    # Simulate a request with prompt_cache_key
-    messages = [
-        {"role": "user", "content": "Hello, how are you?"}
-    ]
-    
-    optional_params = {
-        "max_tokens": 100,
-        "temperature": 0.7,
-        "prompt_cache_key": "test-cache-key-12345",  # This should be removed
-        "stream": True,  # This should also be removed for Bedrock
-    }
-    
-    litellm_params = {}
-    headers = {}
-    
-    # Transform the request
-    result = config.transform_request(
-        model="anthropic.claude-3-7-sonnet-20250219-v1:0",
-        messages=messages,
-        optional_params=optional_params,
-        litellm_params=litellm_params,
-        headers=headers,
+    # Verify that prompt_cache_key is NOT in Bedrock Anthropic's supported params
+    supported_params = config.get_supported_openai_params("anthropic.claude-3-7-sonnet-20250219-v1:0")
+    assert "prompt_cache_key" not in supported_params, (
+        "prompt_cache_key should not be in Bedrock Anthropic's supported params list"
     )
     
-    # Verify prompt_cache_key is NOT in the result
-    assert "prompt_cache_key" not in result, "prompt_cache_key should be removed from Bedrock request"
+    # Verify other common parameters are present
+    assert "max_tokens" in supported_params or "max_completion_tokens" in supported_params
+    assert "temperature" in supported_params
+    assert "tools" in supported_params
     
-    # Verify model is also removed (Bedrock specific)
-    assert "model" not in result, "model should be removed from Bedrock request"
-    
-    # Verify stream is also removed (Bedrock specific)
-    assert "stream" not in result, "stream should be removed from Bedrock request"
-    
-    # Verify anthropic_version is added (Bedrock specific)
-    assert "anthropic_version" in result
-    
-    # Verify other parameters are still present
-    assert "max_tokens" in result
-    assert result["max_tokens"] == 100
-    assert "temperature" in result
-    assert result["temperature"] == 0.7
-    assert "messages" in result
-    
-    print("✅ Test passed: prompt_cache_key successfully removed from Bedrock Anthropic request")
+    print("✅ Test passed: prompt_cache_key is not in Bedrock Anthropic's supported params")

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

@@ -45,40 +45,15 @@ def test_prompt_cache_key_removed_from_vertex_ai_request():
     """
     config = VertexAIAnthropicConfig()
     
-    # Simulate a request with prompt_cache_key
-    messages = [
-        {"role": "user", "content": "Hello, how are you?"}
-    ]
-    
-    optional_params = {
-        "max_tokens": 100,
-        "temperature": 0.7,
-        "prompt_cache_key": "test-cache-key-12345",  # This should be removed
-    }
-    
-    litellm_params = {}
-    headers = {"anthropic-version": "vertex-2023-10-16"}
-    
-    # Transform the request
-    result = config.transform_request(
-        model="claude-3-7-sonnet-20250219",
-        messages=messages,
-        optional_params=optional_params,
-        litellm_params=litellm_params,
-        headers=headers,
+    # Verify that prompt_cache_key is NOT in Vertex AI Anthropic's supported params
+    supported_params = config.get_supported_openai_params("claude-3-7-sonnet-20250219")
+    assert "prompt_cache_key" not in supported_params, (
+        "prompt_cache_key should not be in Vertex AI Anthropic's supported params list"
     )
     
-    # Verify prompt_cache_key is NOT in the result
-    assert "prompt_cache_key" not in result, "prompt_cache_key should be removed from Vertex AI request"
+    # Verify other common parameters are present
+    assert "max_tokens" in supported_params or "max_completion_tokens" in supported_params
+    assert "temperature" in supported_params
+    assert "tools" in supported_params
     
-    # Verify model is also removed (Vertex AI specific)
-    assert "model" not in result, "model should be removed from Vertex AI request"
-    
-    # Verify other parameters are still present
-    assert "max_tokens" in result
-    assert result["max_tokens"] == 100
-    assert "temperature" in result
-    assert result["temperature"] == 0.7
-    assert "messages" in result
-    
-    print("✅ Test passed: prompt_cache_key successfully removed from Vertex AI Anthropic request")
+    print("✅ Test passed: prompt_cache_key is not in Vertex AI Anthropic's supported params")