PrismML, an AI lab founded by Caltech researchers and advised by UC Berkeley's Ion Stoica, unveiled a Qualcomm Snapdragon AR1 Gen 1 build of its 1-bit Bonsai large language model at Snapdragon Summit on September 24, 2026, aiming to deliver full multimodal reasoning on smart glasses without a cloud round-trip.
A 2-billion-parameter vision-language model on a wearable
The Bonsai release is a 2-billion-parameter vision-language model compressed to 1-bit precision, achieving roughly 4x compression while retaining almost all of its performance on standard benchmarks, according to the company. Running locally on Snapdragon AR1 Gen 1 hardware lets glasses interpret camera feeds and hold conversational context without streaming video to a data center, a design PrismML argues is essential for privacy, latency and battery life.
An open-weight bet against cloud-only stacks
PrismML is pitching its lineup as an alternative to closed cloud-based systems from OpenAI, Anthropic and others, focused on maximising the compute already inside consumer hardware. The Snapdragon build extends work the lab has been doing with Replit, Akamai, Macrometa, Nscale and Hugging Face on on-device inference. Related: Anthropic Launches Claude Marketplace With 2,000+ Plugins And Partner Software and Meta Bets $1,300 On A VR Headset And Six New AI-Glasses SKUs.
Reporting based on coverage from TechCrunch.