PrismML, a startup founded by Caltech researchers, has raised $22.25 million in a seed funding round. The company is backed by investors including Khosla Ventures, Cerberus Capital, and Caltech. The startup is led by Babak Hassibi, a professor at Caltech, and counts Databricks co-founder Ion Stoica as an advisor.
The company focuses on compressing large language models to run on PCs and smartphones. Its latest release, Bonsai 2 27B, compresses the Alibaba Qwen3.8 27B model down to 5.9 GB, representing a 9x to 10x reduction in memory. PrismML reports that Bonsai 2 matches 98% of the original model's benchmark scores.
The compression technique utilizes ternary weights, reducing the storage requirement for each weight from 16 bits to three values: +1, -1, or 0. The company's previous model, released in March, achieved 95% benchmark parity and has been downloaded over 11 million times, with additional smaller models reaching 2.6 million downloads.
PrismML plans to apply its compression technology to models in the several-hundred-billion-parameter range within the next few months. The startup aims to enable advanced AI capabilities to run locally on user devices, which proponents suggest will improve privacy and eliminate the need for cloud processing.
