How do I set the degree of "reasoning" when using REASONING models?

Reasoning allows the model to draw connections between different data, make judgments, and weigh the benefit of potential actions. You can control the degree of reasoning used by these models (i.e., how long the model should generate reasoning tokens before producing its final answer) by adjusting the Reasoning Effort parameter. Of note, tokens are expended during this process even if the “thinking” isn’t displayed in the user interface. In other words, reasoning tokens are billed tokens, so using reasoning consumes credits more quickly.

This parameter is available from the Chat Controls:

Scroll down to find the Reasoning Effort setting. You can choose one of these values: low, medium, or high. Type in the word to set the parameter.

To turn reasoning off again for a model tagged as HYBRID REASONING, set the Reasoning Effort parameter back to default by clicking on the “Custom” label.

Didn’t find what you needed? Please reach out to research.computing@dartmouth.edu.