Adding support for reasoning_effort parameter in OpenAI GPT-5 model configurations

✅ ISSUE RESOLVED - Complete Fix Implemented

The issue is that the reasoning_effort parameter for GPT-5 models was being filtered out during configuration processing. This has been completely resolved.

Root Cause Analysis

• ✅ OpenAI Python SDK (v1.93.0) includes reasoning_effort as a valid parameter
• ✅ AutoGen's create_kwargs includes reasoning_effort
• ❌ PROBLEM: The parameter was missing from the Pydantic configuration models
• ❌ PROBLEM: The parameter was dropped from _raw_config and _create_args

Complete Implementation

• Analyze repository structure and identify the issue location
• Confirm reasoning_effort is in OpenAI API parameters
• Reproduce the issue with test script
• Add reasoning_effort to CreateArguments TypedDict in config/init.py
• Add reasoning_effort to CreateArgumentsConfigModel Pydantic model
• Add documentation for the reasoning_effort parameter
• Create focused test to validate the fix
• Run existing tests to ensure no regression
• Manually verify the fix works end-to-end
• Fix linting issues
• Fix pyright errors
• Run security checks (0 vulnerabilities found)

Changes Made

1. Configuration Models Updated (config/__init__.py)

# TypedDict - Added reasoning_effort parameter with documentation
class CreateArguments(TypedDict, total=False):
    # ... existing fields ...
    reasoning_effort: Optional[Literal["minimal", "low", "medium", "high"]]
    """Controls the amount of effort the model uses for reasoning.
    Only applicable to reasoning models like o1 and o3-mini.
    - 'minimal': Fastest response with minimal reasoning
    - 'low': Faster responses with less reasoning
    - 'medium': Balanced reasoning and speed
    - 'high': More thorough reasoning, may take longer"""

# Pydantic Model - Added reasoning_effort parameter with comment
class CreateArgumentsConfigModel(BaseModel):
    # ... existing fields ...
    # Controls the amount of effort the model uses for reasoning (reasoning models only)
    reasoning_effort: Literal["minimal", "low", "medium", "high"] | None = None

2. Comprehensive Test Suite Added

• ✅ Test reasoning_effort preservation in OpenAI and Azure OpenAI clients
• ✅ Test load_component configuration with reasoning_effort
• ✅ Test serialization/deserialization of reasoning_effort
• ✅ Test validation of valid values ("minimal", "low", "medium", "high", None)
• ✅ Test rejection of invalid values with proper ValidationError
• ✅ All pyright errors fixed with appropriate ignore comments

3. End-to-End Validation Results

✅ Original Issue Reproduction Case:

config = {
    "provider": "AzureOpenAIChatCompletionClient", 
    "config": {
        "azure_endpoint": "https://fake-endpoint.openai.azure.com/",
        "azure_deployment": "gpt-5-2025-08-07",
        "api_version": "2025-02-01-preview", 
        "model": "gpt-5",
        "temperature": 1,
        "seed": 42,
        "reasoning_effort": "low",  # NOW WORKS! 🎉
    }
}

model_client = ChatCompletionClient.load_component(config)
# Raw config: {..., 'reasoning_effort': 'low', ...}  ✅
# Create args: {..., 'reasoning_effort': 'low', ...} ✅

✅ All Validation Scenarios Pass:

• ✓ reasoning_effort='minimal' is valid
• ✓ reasoning_effort='low' is valid
• ✓ reasoning_effort='medium' is valid
• ✓ reasoning_effort='high' is valid
• ✓ reasoning_effort=None is valid
• ✓ reasoning_effort='invalid_value' correctly rejected with ValidationError

✅ Both Client Types Supported:

• Works with OpenAIChatCompletionClient
• Works with AzureOpenAIChatCompletionClient
• Works with ChatCompletionClient.load_component()

✅ Code Quality:

• All linting checks pass
• All existing tests pass (no regression)
• 0 pyright errors (all private usage properly ignored)
• 0 security vulnerabilities found
• Proper type annotations with Literal["minimal", "low", "medium", "high"]
• Documentation added explaining the parameter

Impact

This fix enables users to configure reasoning effort levels for GPT-5 models in AutoGen, allowing for control over the computational resources and response quality for reasoning tasks. The parameter is documented to explain its purpose and valid values.

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.