What is variable effort reasoning?

Variable effort reasoning refers to the ability of the models to support three different reasoning levels: low, medium, and high. These levels are configurable in the system prompt by inserting keywords such as 'Reasoning: low'. Increasing the reasoning level causes the model’s average chain-of-thought (CoT) length to increase, allowing for flexible reasoning effort based on the complexity of the task at hand. This capability helps the models adapt to various tasks by adjusting the depth of reasoning used in their responses^[1].