Its really easy to argue against local models because when it comes to quality, you can argue using the tokens/sec. and when it comes to speed, you can argue using the parameter count. This is not compared to the frontier stuff but it is the frontier of last year that now runs on a local machine. It was impossible to do this last year.