In my short testing, Veo 3.1 has the same temporal consistency problems as Veo 3. It seems to be a later training snapshot of the same model, at best.

What does temporal consistency mean? Which model do you think is the best right now?

Try to make a human do a flip of any kind. Limbs will magically go through torsos, limbs will swap with each other, and extra limbs will appear and disappear when others are occluded.