Variance is the degree to which a model’s learned function changes in response to differences in training data. High-variance models can fit training data closely but may overfit and perform poorly on new data.