The model's estimate of how likely each label is. predict() turns it into a label using a cut-off of 0.5 unless told otherwise.

ExampleA month-to-month fibre customer scores 0.68 'likely to leave'; the retention team might call the top 100 scores whatever the cut-off.