PrismML brings its tiny LLMs to Qualcomm-powered smart glasses

PrismML, an AI lab launched by Caltech researchers with guidance from UC Berkeley’s Ion Stoica, presented a compact language model designed for Qualcomm’s wearable chipset at the Snapdragon Summit on September 24, 2026. The demonstration centered on a 1-bit Bonsai LLM adapted to run locally on devices using the Snapdragon AR1 Gen 1 platform.

According to the company and previous reporting by TechCrunch, PrismML’s approach compresses larger models by about fourfold while preserving nearly the same benchmark performance. The smart-glasses variant is a 2-billion-parameter model that has been tuned for both vision and language tasks, enabling wearers to query what they see in real time without sending data to remote servers.

PrismML frames this work as part of a broader push for open-weight AI that executes on-device, aiming to make better use of existing endpoint compute and to offer an alternative to cloud-dependent proprietary labs. Running inference locally can reduce network and server reliance and, in PrismML’s view, lessen exposure to third-party data handling.

While showcasing the model on Qualcomm’s AR platform is a practical step toward wider hardware support, PrismML and Qualcomm have not announced any consumer smart glasses that will ship with the model preinstalled. The demonstration, however, signals growing interest in fitting capable language-and-vision models into compact wearables.