A model's learned ability to choose between different reasoning strategies (e.g., quick vs. extended) based on problem characteristics.