remove unsupported prompt_cache_key for anthropic models

litellm/llms/anthropic/chat/transformation.py

@@ -736,6 +736,9 @@ 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,

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

@@ -80,6 +80,7 @@ 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

@@ -128,8 +128,12 @@ class AmazonAnthropicClaudeMessagesConfig(
         # 3. `model` is not allowed in request body for bedrock invoke
         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)
             
-        # 4. Handle anthropic_beta from user headers
+        # 5. 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/vertex_ai/vertex_ai_partner_models/anthropic/transformation.py

@@ -68,6 +68,7 @@ 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(

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

@@ -364,3 +364,50 @@ def test_get_supported_params_thinking():
     config = AnthropicConfig()
     params = config.get_supported_openai_params(model="claude-sonnet-4-20250514")
     assert "thinking" in params
+
+
+def test_prompt_cache_key_removed_from_request():
+    """
+    Test that prompt_cache_key is removed from the request before sending to Anthropic.
+    
+    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 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 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")

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

@@ -20,3 +20,60 @@ def test_get_supported_params_thinking():
         model="anthropic.claude-sonnet-4-20250514-v1:0"
     )
     assert "thinking" in params
+
+
+def test_prompt_cache_key_removed_from_bedrock_request():
+    """
+    Test that prompt_cache_key is removed from Bedrock Anthropic requests.
+    
+    This prevents the error: "prompt_cache_key: Extra inputs are not permitted"
+    when using Claude models on AWS Bedrock.
+    
+    Related issue: Bedrock doesn't support prompt_cache_key parameter
+    """
+    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 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 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")

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

@@ -32,3 +32,53 @@ def test_get_supported_params_thinking():
     config = VertexAIAnthropicConfig()
     params = config.get_supported_openai_params(model="claude-sonnet-4")
     assert "thinking" in params
+
+
+def test_prompt_cache_key_removed_from_vertex_ai_request():
+    """
+    Test that prompt_cache_key is removed from Vertex AI Anthropic requests.
+    
+    This prevents the error: "prompt_cache_key: Extra inputs are not permitted"
+    when using Claude models on Vertex AI.
+    
+    Related issue: Vertex AI doesn't support prompt_cache_key parameter
+    """
+    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 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 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")