Where are you getting these numbers from? I'm finding data between 0.3Wh and 3Wh for a 4o query, while a typical phone battery has 10Wh.
The energy consumption per prompt will decrease for the same task complexity, as inference hardware gets better (as they already have from the 2023 paper that reported 3Wh).
The energy consumption per prompt will decrease for the same task complexity, as inference hardware gets better (as they already have from the 2023 paper that reported 3Wh).