PrismML launches Bonsai 2 27B reasoning model with 9-10x memory reduction

AI startup PrismML released Bonsai 2 27B on Thursday, its latest reasoning model that compresses Alibaba's open-source Qwen3.8 27B down to 5.9GB. The company says the model is small enough to fit on a PC and possibly a high-end smartphone, with a 9x to 10x reduction in memory footprint compared with the original.

According to PrismML, Bonsai 2 achieves 98% of Qwen's aggregate benchmark scores, up from 95% for the first-generation Bonsai released in March 2026. The company said the first Bonsai model has been downloaded more than 11 million times, and its smaller models have been downloaded another 2.6 million times.

PrismML was founded by Caltech researchers and is led by Babak Hassibi, a Caltech professor specializing in compression technologies. The startup has raised a $22.25 million seed round from investors including Khosla Ventures, Cerberus Capital and Caltech.

The company's compression approach uses "ternary" weights, reducing each weight from 16 bits to one of three values: +1, -1 or 0. PrismML says it plans to apply the technique to larger models in the hundreds of billions of parameters in the coming months.