It is always equal to or larger than R-squared.

It penalizes the addition of unnecessary variables, preventing an artificial increase in goodness-of-fit metrics.

It does not rely on the assumption of normality of errors.

It is required for testing the statistical significance of individual coefficients.