SAGE-PTQ is a new ultra-low-bit quantization framework for large language models that significantly reduces hidden scaling costs and improves efficiency. It outperforms existing methods, achieving faster decoding and lower memory usage on models like LLaMA-3-8B and LLaMA-2-70B.