Models that support reasoning features, such as Claude, usually offer multiple levels of reasoning effort settings, including low, medium, high, extra high, and max, and the higher the level, the higher the token consumption and cost.