PrismML Compresses Qwen 27B to 5.9GB With 98% Benchmark Retention

PrismML released Bonsai 2 27B, a ternary-quantized compression of Alibaba's open-source Qwen3.8 27B. The 5.9GB model uses +1, -1, 0 weights versus 16-bit precision, cutting memory 9x to 10x. PrismML reports 98% aggregate benchmark parity, up from 95% for its March Bonsai release downloaded 11 million times.