fix: add store to OPENAI_CHAT_COMPLETION_PARAMS (#21195)

The OpenAI `store` parameter (used for storing completions for
distillation/evals) was missing from `OPENAI_CHAT_COMPLETION_PARAMS`.

This caused it to be unrecognized by `get_standard_openai_params()` and
the `litellm_proxy` provider config. It also meant that code paths using
this list (rather than `DEFAULT_CHAT_COMPLETION_PARAM_VALUES`) would
treat `store` as a provider-specific parameter and forward it to
non-OpenAI providers like Anthropic, resulting in:

    "store: Extra inputs are not permitted"

Fixes #19700

litellm/constants.py

@@ -576,6 +576,7 @@ OPENAI_CHAT_COMPLETION_PARAMS = [
     "thinking",
     "web_search_options",
     "service_tier",
+    "store",
 ]
 
 OPENAI_TRANSCRIPTION_PARAMS = [

tests/llm_translation/test_optional_params.py

@@ -1894,28 +1894,28 @@ def test_validate_openai_optional_params_stop_truncation():
     result = validate_openai_optional_params(stop=stop_sequences)
     assert result == ["stop1", "stop2", "stop3", "stop4"]
     assert len(result) == 4
-    
+
     # Test with exactly 4 stop sequences - should not truncate
     stop_sequences_4 = ["stop1", "stop2", "stop3", "stop4"]
     result = validate_openai_optional_params(stop=stop_sequences_4)
     assert result == ["stop1", "stop2", "stop3", "stop4"]
     assert len(result) == 4
-    
+
     # Test with less than 4 stop sequences - should not truncate
     stop_sequences_2 = ["stop1", "stop2"]
     result = validate_openai_optional_params(stop=stop_sequences_2)
     assert result == ["stop1", "stop2"]
     assert len(result) == 2
-    
+
     # Test with single stop sequence as string - should return as is
     stop_string = "stop1"
     result = validate_openai_optional_params(stop=stop_string)
     assert result == "stop1"
-    
+
     # Test with None - should return None
     result = validate_openai_optional_params(stop=None)
     assert result is None
-    
+
     # Test with empty list - should return empty list
     result = validate_openai_optional_params(stop=[])
     assert result == []
@@ -1928,7 +1928,7 @@ def test_validate_openai_optional_params_disable_stop_sequence_limit():
     """
     # Save original value
     original_value = litellm.disable_stop_sequence_limit
-    
+
     try:
         # Test with disable_stop_sequence_limit = True - should NOT truncate
         litellm.disable_stop_sequence_limit = True
@@ -1936,7 +1936,7 @@ def test_validate_openai_optional_params_disable_stop_sequence_limit():
         result = validate_openai_optional_params(stop=stop_sequences)
         assert result == ["stop1", "stop2", "stop3", "stop4", "stop5", "stop6"]
         assert len(result) == 6
-        
+
         # Test with disable_stop_sequence_limit = False - should truncate to 4
         litellm.disable_stop_sequence_limit = False
         stop_sequences = ["stop1", "stop2", "stop3", "stop4", "stop5", "stop6"]
@@ -1965,19 +1965,83 @@ def test_validate_openai_optional_params_integration():
             mock_response.usage.prompt_tokens = 10
             mock_response.usage.completion_tokens = 5
             mock_response.usage.total_tokens = 15
-            
-            mock_client.return_value.chat.completions.create.return_value = mock_response
-            
+
+            mock_client.return_value.chat.completions.create.return_value = (
+                mock_response
+            )
+
             # Call completion with more than 4 stop sequences
             response = litellm.completion(
                 model="gpt-3.5-turbo",
                 messages=[{"role": "user", "content": "Hello"}],
                 stop=["stop1", "stop2", "stop3", "stop4", "stop5", "stop6"],
-                mock_response="Test response"  # This will use mock
+                mock_response="Test response",  # This will use mock
             )
-            
+
             # Verify the call was made (stop sequences should be truncated internally)
             assert response is not None
     except Exception as e:
         # Should not raise an exception
         pytest.fail(f"validate_openai_optional_params integration failed: {e}")
+
+
+def test_drop_store_param_for_anthropic():
+    """
+    Test that the OpenAI-specific `store` parameter is correctly dropped
+    when calling Anthropic with drop_params=True.
+
+    `store` is an OpenAI Chat Completion parameter (for storing completions
+    for distillation/evals) that Anthropic does not support. Without proper
+    handling, it leaks through to the Anthropic API and causes a
+    "store: Extra inputs are not permitted" error.
+
+    Ref: https://github.com/BerriAI/litellm/issues/19700
+    """
+    optional_params = get_optional_params(
+        model="claude-sonnet-4-20250514",
+        custom_llm_provider="anthropic",
+        drop_params=True,
+        store=True,
+    )
+    assert "store" not in optional_params
+
+
+def test_additional_drop_params_store_for_anthropic():
+    """
+    Test that `additional_drop_params=["store"]` correctly strips the `store`
+    parameter for non-OpenAI providers like Anthropic.
+
+    Ref: https://github.com/BerriAI/litellm/issues/19700
+    """
+    optional_params = get_optional_params(
+        model="claude-sonnet-4-20250514",
+        custom_llm_provider="anthropic",
+        additional_drop_params=["store"],
+        store=True,
+    )
+    assert "store" not in optional_params
+
+
+def test_store_in_openai_chat_completion_params():
+    """
+    Test that `store` is recognized as a standard OpenAI Chat Completion
+    parameter. This ensures it is correctly handled by helper functions
+    like `get_standard_openai_params()` and provider configs that rely on
+    `OPENAI_CHAT_COMPLETION_PARAMS`.
+
+    Without `store` in this list, functions that filter by known OpenAI
+    params will silently drop it for OpenAI calls or incorrectly treat
+    it as a provider-specific param for non-OpenAI providers.
+
+    Ref: https://github.com/BerriAI/litellm/issues/19700
+    """
+    from litellm.constants import OPENAI_CHAT_COMPLETION_PARAMS
+
+    assert "store" in OPENAI_CHAT_COMPLETION_PARAMS
+
+    # Verify get_standard_openai_params recognizes store
+    from litellm.utils import get_standard_openai_params
+
+    result = get_standard_openai_params({"store": True, "temperature": 0.7})
+    assert "store" in result
+    assert result["store"] is True