This can help you evaluate agentic capabilites of various llm's. How they explore codebase.