To clarify, it is not the total number of words but rather the number of unique words considered. Imho a million of unique words is okay. A bigger concern for me would be that words on Wikipedia can be overly specific.
For this metric, Wikipedia might not be a representative dataset. Wikipedia uses many technical terms and composite words which tend to be longer than words that are more common in an everyday dialect.
There were some comments on Reddit suggesting that cutting the dataset at 15 and removing 40% of words was not the best move. I have locally built a version with the limit set to 30.
Oh, somehow I missed the theme setting. I tried both firefox and chromium and got the light theme by default despite having my system/browser settings set to prefer a dark theme.