Fix gpt-5-codex with Codex CLI: Resolve negative reasoning IDs and missing tool calls

litellm/completion_extras/litellm_responses_transformation/transformation.py

@@ -73,6 +73,23 @@ def _handle_raw_dict_response_item(
         if item_type == "reasoning":
             return None, index
 
+        # Handle function_call items (for gpt-5-codex and similar models)
+        # Issue: https://github.com/BerriAI/litellm/issues/14846
+        if item_type == "function_call":
+            msg = Message(
+                content=None,
+                tool_calls=[{
+                    "id": item.get("call_id"),
+                    "function": {
+                        "name": item.get("name"),
+                        "arguments": item.get("arguments", ""),
+                    },
+                    "type": "function",
+                }]
+            )
+            choice = Choices(message=msg, finish_reason="tool_calls", index=index)
+            return choice, index + 1
+
         # Handle message items with output_text content
         if item_type == "message":
             content_list = item.get("content", [])

litellm/llms/openai/responses/transformation.py

@@ -124,7 +124,9 @@ def _handle_reasoning_item(self, item: Dict[str, Any]) -> Dict[str, Any]:
                 # Ensure required fields are present for ResponseReasoningItem
                 item_data = dict(item)
                 if "id" not in item_data:
-                    item_data["id"] = f"rs_{hash(str(item_data))}"
+                    # Use abs() to ensure ID is always positive (hash() can return negative values)
+                    # Issue: https://github.com/BerriAI/litellm/issues/14846
+                    item_data["id"] = f"rs_{abs(hash(str(item_data)))}"
                 if "summary" not in item_data:
                     item_data["summary"] = (
                         item_data.get("reasoning_content", "")[:100] + "..."

tests/test_litellm/llms/openai/responses/test_openai_responses_transformation.py

@@ -775,4 +775,160 @@ def test_get_supported_openai_params():
     assert "temperature" in params
     assert "stream" in params
     assert "background" in params
-    assert "stream" in params
+    assert "stream" in params
+
+
+def test_reasoning_id_generation_always_positive():
+    """
+    Test that reasoning IDs generated by _handle_reasoning_item are always positive.
+    Bug fix for issue #14846 - hash() can return negative values causing OpenAI API errors.
+    """
+    config = OpenAIResponsesAPIConfig()
+
+    # Test multiple iterations to increase chance of catching negative hash() values
+    for i in range(100):
+        test_item = {
+            "type": "reasoning",
+            "summary": [{"type": "summary_text", "text": f"test reasoning {i}"}]
+            # No ID provided - will be generated
+        }
+
+        result = config._handle_reasoning_item(test_item)
+
+        # Verify ID exists and is properly formatted
+        assert "id" in result, f"Iteration {i}: ID should be generated"
+        assert result["id"].startswith("rs_"), f"Iteration {i}: ID should start with 'rs_'"
+
+        # Critical assertion: ID should NEVER contain negative sign
+        assert "rs_-" not in result["id"], f"Iteration {i}: ID should not contain negative number: {result['id']}"
+
+        # Verify ID is numeric after 'rs_' prefix
+        id_value = result["id"][3:]  # Remove 'rs_' prefix
+        assert id_value.isdigit(), f"Iteration {i}: ID value should be numeric: {result['id']}"
+
+
+def test_reasoning_id_generation_with_existing_id():
+    """Test that existing reasoning IDs are preserved and not regenerated."""
+    config = OpenAIResponsesAPIConfig()
+
+    existing_id = "rs_12345678"
+    test_item = {
+        "type": "reasoning",
+        "id": existing_id,
+        "summary": [{"type": "summary_text", "text": "test"}]
+    }
+
+    result = config._handle_reasoning_item(test_item)
+
+    # Existing ID should be preserved
+    assert result["id"] == existing_id
+
+
+def test_handle_raw_dict_response_function_call():
+    """
+    Test that function_call items from Responses API are properly handled.
+    Bug fix for issue #14846 - gpt-5-codex tool calls were being ignored.
+    """
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        LiteLLMResponsesTransformationHandler,
+    )
+    from litellm.types.utils import Choices
+
+    handler = LiteLLMResponsesTransformationHandler()
+
+    # Simulate a function_call item as returned by gpt-5-codex
+    function_call_item = {
+        "type": "function_call",
+        "id": "fc_12345",
+        "call_id": "call_abcdef",
+        "name": "shell",
+        "arguments": '{"command": ["ls", "-la"]}',
+        "status": "completed"
+    }
+
+    choice, new_index = handler._handle_raw_dict_response_item(function_call_item, 0)
+
+    # Verify choice was created (not None)
+    assert choice is not None, "function_call items should return a Choice object"
+    assert isinstance(choice, Choices), "Result should be a Choices object"
+
+    # Verify index was incremented
+    assert new_index == 1, "Index should be incremented after handling function_call"
+
+    # Verify the choice has correct structure
+    assert choice.finish_reason == "tool_calls", "finish_reason should be 'tool_calls'"
+    assert choice.index == 0, "Choice index should match input index"
+
+    # Verify message contains tool_calls
+    assert choice.message is not None, "Message should not be None"
+    assert choice.message.content is None, "Content should be None for tool calls"
+    assert choice.message.tool_calls is not None, "tool_calls should not be None"
+    assert len(choice.message.tool_calls) == 1, "Should have one tool call"
+
+    # Verify tool call structure
+    tool_call = choice.message.tool_calls[0]
+    assert tool_call["id"] == "call_abcdef", "Tool call ID should match call_id"
+    assert tool_call["type"] == "function", "Tool call type should be 'function'"
+    assert tool_call["function"]["name"] == "shell", "Function name should match"
+    assert tool_call["function"]["arguments"] == '{"command": ["ls", "-la"]}', "Arguments should match"
+
+
+def test_handle_raw_dict_response_multiple_function_calls():
+    """Test handling multiple function_call items in sequence."""
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        LiteLLMResponsesTransformationHandler,
+    )
+
+    handler = LiteLLMResponsesTransformationHandler()
+
+    function_calls = [
+        {
+            "type": "function_call",
+            "call_id": "call_1",
+            "name": "shell",
+            "arguments": '{"command": ["ls"]}'
+        },
+        {
+            "type": "function_call",
+            "call_id": "call_2",
+            "name": "read_file",
+            "arguments": '{"path": "test.txt"}'
+        }
+    ]
+
+    index = 0
+    choices = []
+
+    for fc in function_calls:
+        choice, index = handler._handle_raw_dict_response_item(fc, index)
+        if choice is not None:
+            choices.append(choice)
+
+    # Verify both function calls were handled
+    assert len(choices) == 2, "Should have two choices for two function calls"
+    assert index == 2, "Index should be incremented for each function call"
+
+    # Verify each choice has correct tool call
+    assert choices[0].message.tool_calls[0]["function"]["name"] == "shell"
+    assert choices[1].message.tool_calls[0]["function"]["name"] == "read_file"
+
+
+def test_handle_raw_dict_response_reasoning_ignored():
+    """Test that reasoning items are still ignored (no change to existing behavior)."""
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        LiteLLMResponsesTransformationHandler,
+    )
+
+    handler = LiteLLMResponsesTransformationHandler()
+
+    reasoning_item = {
+        "type": "reasoning",
+        "id": "rs_12345",
+        "summary": [{"type": "summary_text", "text": "test reasoning"}]
+    }
+
+    choice, new_index = handler._handle_raw_dict_response_item(reasoning_item, 0)
+
+    # Reasoning items should still return None
+    assert choice is None, "Reasoning items should return None"
+    assert new_index == 0, "Index should not change for reasoning items"