The AI Lab PrismML has released a version of its Bonsai LLM for smart glasses powered by Qualcomm’s Snapdragon AR1 Gen 1 Platform. Demonstrated at the Snapdragon Summit, the 1-bit model runs locally on the device without needing cloud connectivity. It shrinks a standard model size by four times while keeping performance on benchmarks almost identical. The specific implementation uses a 2-billion-parameter architecture tuned for vision and language tasks. This allows wearers to ask questions about objects in their field of view and receive immediate answers directly from the hardware.
This move prioritises running open-weight models on existing device silicon rather than relying on proprietary cloud services. The approach challenges the industry habit of sending data to external servers where privacy concerns often arise. By compressing models to run on chips designed for augmented reality, the startup aims to reduce dependency on massive compute clusters. It also removes the need for constant internet access during use.
- The model compresses standard weights by a factor of four.
- It targets the Snapdragon AR1 Gen 1 Platform specifically.
- No commercial smart glasses running this software have been announced yet.
