PrismML compresses AI models to run on phones and PCs
Startup PrismML has released Bonsai 2 27B, a compressed language model that fits in just 5.9 GB while retaining nearly all the performance of its original. The company aims to let powerful AI run directly on personal devices instead of the cloud.
AI startup PrismML unveiled Bonsai 2 27B on Thursday, the latest release in its family of compressed language models. The model shrinks Qwen3.8 27B, a widely used open-source model from Alibaba, down to just 5.9 GB — a nine- to tenfold reduction from the original's size. That footprint is small enough to run on an ordinary PC and potentially on a high-end smartphone.
The company was founded by a group of Caltech researchers and is led by CEO Babak Hassibi, a Caltech professor specializing in compression technologies. Ion Stoica, co-founder of Databricks and director of UC Berkeley's Sky Computing Lab, serves as an advisor. PrismML is backed by Khosla Ventures, Cerberus Capital, and Caltech, and has raised a $22.25 million seed round so far.
According to Hassibi, PrismML's compression approach stands out because it causes almost no loss in performance compared to the original models. Bonsai 2 matches 98% of Qwen's aggregate benchmark scores, an improvement over the first Bonsai model released in March, which matched 95%. That original model has already been downloaded more than 11 million times, while PrismML's newer, smaller models have added another 2.6 million downloads.
The technique relies on what the company calls "ternary" weights. Normally, each weight in a model requires 16 bits to store; PrismML's method reduces this to just three possible values — +1, −1, or 0 — dramatically cutting the space a model needs.
PrismML's next goal is to apply this compression method to much larger models, in the range of several hundred billion parameters, with a release expected within the next couple of months. Hassibi believes larger models will actually be easier to compress without losing capability. Stoica says the technology could make advanced AI available for free and with greater privacy, since it would run directly on users' own devices rather than sending data to the cloud.
