PrismML hopes its tiny LLM will change how we all use AI

PrismML released Bonsai 2 27B, a compressed version of Qwen3.8 27B reduced to 5.9 GB. The model targets local execution on PCs and high-end smartphones. PrismML reports the release matches 98% of the original benchmark scores, improving upon the previous version's 95% alignment. This update enables developers to deploy reasoning models with significantly lower memory requirements.