Developing2 outlets · over 18 hours
PrismML Launches Compressed Bonsai 2 AI Model
AI lab PrismML released Bonsai 2 27B, compressing Alibaba's Qwen3.8 27B model to 5.9 GB for local use on consumer hardware. The startup achieved a 9x to 10x memory reduction while retaining 98% of the original model's benchmark performance.
The story so far2 updates, latest first.
Tracked by ByteBrief