How would you describe them?
A quantization compiler that delivers 12× lower KL than Q4_K_M at the same file size