A quantized variant of the GLM family, this model trades some precision for reduced memory footprint by mixing Int4 and Int8 quantization strategies. It handles text-in, text-out tasks and brings an unusually large context window of over one million tokens to lower-resource environments. The trade-off is that quantization can introduce subtle quality degradation compared to full-precision counterparts.