3 ms·

I would also like to hear how you benchmarked the accuracy. Comparing output between competing GPT models?