It's been established that it moves in the direction of acceleration, but an addendum I thought of is that the balloon's speed will at first keep getting faster as the gradient is built up as the air is squished to one side, then slow down as the air pressure reaches equilibrium again.