A technique where a model allocates more computational resources and time during inference (when generating answers) to improve quality and accuracy on harder problems.
Multi-step reasoning, logic puzzles, mathematical problem-solving