2 ms·

I assume the training dataset is mostly the same anyway. I imagine prompting in different language could have a huge effect though.