3 ms·The model has a new improved tokenizer that also counts the same outputs as 20% more tokensby htrp 2mo agoThe model has a new improved tokenizer that also counts the same outputs as 20% more tokens