The number of bits used to represent each number in a model; lower bit precision (like 3-bit) means smaller file size but potentially less accurate calculations.