PrismML shows a 1‑bit 2B model running on Qualcomm's smart‑glasses chip
The Caltech-founded startup says its compressed Bonsai model is a quarter of the size and keeps nearly all its benchmark scores, though no glasses ship with it yet.
PrismML demonstrated a 1-bit version of its 2-billion-parameter Bonsai model running locally on Qualcomm's Snapdragon AR1 Gen 1 platform, the chip inside several smart-glasses designs. The demonstration took place at Qualcomm's Snapdragon Summit on Wednesday, TechCrunch reported on Thursday. No other outlet has the story yet.
The compression claim is the company's. PrismML said the 1-bit build is a quarter of the size of the original and retains nearly all of its performance on benchmarks. It did not name those benchmarks, and TechCrunch's report carries no measurement of its own.
The model is tuned for vision and language tasks, which is what glasses ask of a model: read the scene, answer a question about it. Caltech researchers founded PrismML, and Ion Stoica of the University of California, Berkeley, advises it, according to TechCrunch.
No product exists yet. TechCrunch said no smart glasses carry PrismML's models so far, and gave the company's stated goal as open-weight AI that runs on devices and makes better use of the computing power those devices already have.