3 ms·

I realized the same with the GPT3.5 model too. After the Last update it started giving shorter and less coherent answers.