PrismML Brings Tiny Bonsai LLM to Qualcomm Smart Glasses Platform

PrismML has adapted its tiny language models for AI smart glasses powered by Qualcomm’s Snapdragon chips, bringing its compact, on-device AI approach to wearable hardware.

At Qualcomm’s Snapdragon Summit, the chipmaker demonstrated PrismML’s 1-bit Bonsai LLM running locally on AI smart glasses built on the Snapdragon AR1 Gen 1 Platform.

PrismML, an AI lab founded by Caltech researchers and advised by UC Berkeley’s Ion Stoica, develops smaller language models designed to preserve most of the performance of larger models while requiring fewer computing resources.

For smart glasses, PrismML created a 2-billion-parameter Bonsai model tuned for both vision and language. The company says its approach reduces the model’s size by four times while retaining nearly all of its performance on standard benchmarks.

The vision-language model is designed to let users ask questions about what they are seeing in real time, with the AI processing those requests directly on the device.

Running the model locally fits PrismML’s broader goal of developing open-weight AI that can make better use of computing power already available on consumer devices. The company positions that approach as an alternative to relying on proprietary AI services and the additional computing infrastructure required to run models remotely.

Support for Qualcomm’s Snapdragon AR1 Gen 1 Platform moves PrismML closer to that goal, but no commercially announced pair of smart glasses yet uses the company’s Bonsai model.