I don't really care. The model may be competitive, but my use cases require speed, local (semi-local) execution and reliability. Neither of these seem to be bak

I don't really care. The model may be competitive, but my use cases require speed, local (semi-local) execution and reliability. Neither of these seem to be baked into whatever X produced now. When they make the mini model available for download and quantizable. That's when I may be interested. But given the minimal improvement in the past several months, I'm inclined to believe that we have reached the plateau.