PrismML announced at the Snapdragon Summit 2026 that its 1-bit Bonsai vision-language model can run locally on AI smart glasses powered by the Qualcomm Snapdragon AR1 Gen 1 Platform.
PrismML's 1-Bit Bonsai Models to Run Locally on Snapdragon-Powered Smart Glasses
The new model is a 2-billion-parameter vision-language model built on PrismML's Bonsai 1.7B. By using 1-bit and 3-bit precision to reduce memory usage, the technology allows for significantly more intelligence within the same hardware constraints. On the Snapdragon AR1 Gen 1 platform, the 1-bit 2B Bonsai model provides intelligence equivalent to a 4-bit precision model while using approximately 4x less memory and generating tokens at more than 2x the speed.
PrismML optimized the model's weights and architecture specifically for the Qualcomm Hexagon NPU to maximize efficiency. According to PrismML, this optimization is critical for smart glasses, which must operate within tight limits for memory, power, and thermal management while remaining responsive throughout the day.
The technology enables various real-time experiences, such as understanding what the wearer sees and reasoning over that visual context. Ziad Asghar, Qualcomm's Senior Vice President and General Manager of XR, Wearables and Personal AI, stated that supporting PrismML’s 1-bit models helps make advanced AI experiences more practical on lightweight, power-optimized wearables.
Sources
- 小型AIモデル「1-Bit Bonsai」シリーズがスマートグラスで動作するように、Qualcommとの協力により （GIGAZINE, 2026-09-25）
- PrismML