>so they place much more emphasis on compute-efficiency. Maybe on training, but on inference they use more tokens than comparable western models. https://arti

>so they place much more emphasis on compute-efficiency. Maybe on training, but on inference they use more tokens than comparable western models. https://artificialanalysis.ai/?output-tokens=intelligence-vs-output-tokens https://artificialanalysis.ai/?output-tokens=intelligence-vs...