3 ms·

Then why is cheaper per token, and following the same path as GPT3 with several cheaper turbo versions.