PrismML, tiny LLMs debut on Qualcomm smart glasses

Original: PrismML brings its tiny LLMs to Qualcomm-powered smart glasses

Why This Matters

On-device LLMs for wearables signal a real shift toward privacy-first AI at the edge, with Qualcomm's AR platform as the proving ground.

PrismML, a Caltech-founded AI startup advised by UC Berkeley's Ion Stoica, demonstrated its 1-bit Bonsai LLM on Qualcomm's Snapdragon AR1 Gen 1 Platform at Snapdragon Summit on September 24, 2026. The 2-billion-parameter model runs locally on smart glasses for real-time vision and language queries.

PrismML's core technology compresses larger models by 4x while retaining near-full performance on standard benchmarks. The Bonsai LLM shown at Qualcomm's Snapdragon Summit is a 2B-parameter model optimized for vision and language tasks — letting wearers ask questions about what they see in real time, with all processing happening on-device.

The startup, founded by Caltech researchers and backed by Ion Stoica of UC Berkeley fame, frames its work as an explicit counter to proprietary AI labs: local inference means no data leaves the device, no privacy promises required, and no dependency on cloud compute. Running on Qualcomm's Snapdragon AR1 Gen 1 chip moves that argument from theory to hardware.

Still, the showcase remains a proof of concept. No smart glasses product running PrismML has been announced by any manufacturer. Qualcomm featured the demo, but a shipping device with PrismML inside is still an open question.

Source

techcrunch.com — Read original →