A model dial (high / medium / low) for how hard the model thinks before responding, set separately from which model and how fast it runs.