If we’re talking about AI performance the software is there, and models often get ported to MLX.