At Qualcomm's Snapdragon Summit on Wednesday, AI startup PrismML showcased its 1-bit Bonsai LLM, designed to run locally on AI smart glasses powered by the Snapdragon AR1 Gen 1 Platform.
PLUS ULTRAModel ReleasesPrismMLBonsai LLM
PrismML adapts tiny LLMs for Qualcomm-powered smart glasses
PLUS ULTRA by Amenoyomi
PrismML, founded by Caltech researchers and advised by UC Berkeley’s Ion Stoica, specializes in shrinking large models—by as much as 4x in this instance—while maintaining performance on standard benchmarks. The specific version for smart glasses is a 2-billion-parameter model tuned for vision and language, intended to enable users to ask questions about their surroundings in real time.
The startup aims to advance open-weight AI that utilizes existing on-device computing power, offering an alternative to proprietary AI labs that rely heavily on massive compute resources and privacy promises. While the model is now available for Qualcomm's chips, no smart glasses featuring PrismML have been officially announced yet.
PLUS ULTRAby Amenoyomi
The 1-bit Bonsai LLM enables local execution by compressing larger models by four times while retaining nearly all performance on standard benchmarks. This reduction allows a 2-billion-parameter model, tuned for vision and language, to operate on hardware such as the Snapdragon AR1 Gen 1 Platform without relying on external servers.
This on-device capability is designed as an alternative to the privacy promises of proprietary AI labs, as it keeps data processing local to the device. Furthermore, by leveraging the existing computing power within devices, the approach seeks to reduce the dependence on the massive, centralized compute resources typically required for the development and operation of proprietary models.
Sources
- PrismML brings its tiny LLMs to Qualcomm-powered smart glasses （TechCrunch AI, 2026-09-24）