Their sales pitch seems to be roughly that the value is more in the machine learning software than the gripper itself. I assume having both means the software can choose how much of each (fingers/suction) to use for a particular item, after it identifies the item.

So perhaps it's supposed to work best when there's a variety of items versus a few predictable ones.

This article has some background: https://xconomy.com/boston/2017/03/28/righthand-robotics-pic...

Also, they seem to have started with fingers only, so there's probably some bias to keep them.