The Pulse
PrismML Puts 1-Bit Bonsai on Snapdragon Smart Glasses
PrismML’s 2-billion-parameter vision-language model ran locally on AI glasses powered by Qualcomm’s Snapdragon AR1 Gen 1, the company says, with Qualcomm tests reporting lower weight memory use and faster token generation for the 1-bit lang
AI.info Team · Written with AI · fact-checked ·
“For AI to become ubiquitous, capable intelligence has to fit into the devices people already carry and wear. Smart glasses make that challenge very concrete: you have an extraordinary amount of context available through vision, but an extremely constrained device in which to understand it. Working with Qualcomm Technologies, we designed Bonsai around those constraints. The result is substantially more intelligence within the same hardware envelope. We believe that model-hardware co-design is how AI moves from the cloud into the devices people use every day.”
Babak Hassibi, co-founder and CEO of PrismML
PrismML demonstrates 1-bit Bonsai on Snapdragon glasses
PrismML announced on September 23, 2026, that its 1-bit Bonsai model had been demonstrated running locally on AI smart glasses powered by Qualcomm’s Snapdragon AR1 Gen 1 platform. The demonstration took place during the Snapdragon Summit. The model can fit four times as many parameters as a model that previously could run within the same memory constraints on some glasses form factors, PrismML says.
The demonstrated model is a 2-billion-parameter vision-language model built on PrismML’s Bonsai 1.7B. It combines a 1.7-billion-parameter 1-bit language model with a 0.3-billion-parameter 4-bit vision encoder. PrismML worked with Qualcomm Technologies to optimize the model’s weights and architecture for Qualcomm’s Hexagon neural processing unit. The company describes possible tasks as understanding what a wearer sees, reasoning about that context and responding in real time.
Qualcomm reports memory and speed results
Qualcomm testing on the AR1 Gen 1 platform found that the 1-bit language model used 0.43 gigabytes for its weights, compared with 1.66 gigabytes for the corresponding 4-bit language model. That is a 3.83-fold reduction, or 74% less weight memory. The figures concern language-model weights, not the total memory use of a complete glasses system. The test used a platform configured with 4GB of memory and a 1,024-token context length.
In the same test setup, the 1-bit model generated 15.36 tokens per second, compared with 7.44 tokens per second for the 4-bit model, a 2.06-fold improvement. Qualcomm compiled the model using an internal software development kit with 1-bit kernel support. Qualcomm’s testing evaluated the Bonsai 2B vision-language model, including its 1.7B 1-bit language model and 0.3B 4-bit vision encoder; the weight-memory and token-generation comparisons were between the 1-bit and corresponding 4-bit language models.
PrismML separately reports benchmark results comparable to a 4-bit Qwen 3 1.7B model. Those results came from PrismML’s September 2026 evaluations, which compared the Bonsai 1.7B 1-bit language model with a corresponding 4-bit Qwen 3 1.7B model. They are distinct from Qualcomm’s hardware measurements on the Snapdragon platform.
“The future of Personal AI depends on bringing powerful intelligence directly to the devices people use every day. By enabling support for PrismML’s innovative 1-bit models on Snapdragon AR1 Gen 1, we’re helping make advanced AI experiences more practical on lightweight, power-optimized wearables. This breakthrough opens the door to richer multimodal AI experiences such as vision-based assistance that run locally, with greater responsiveness and privacy-focused features.”
Ziad Asghar, Senior Vice President and General Manager, XR, Wearables and Personal AI, Qualcomm Technologies, Inc.
Local execution can make interactions more immediate while keeping context on the device, PrismML says. The company and Qualcomm Technologies plan to continue working together on optimizing increasingly capable Bonsai models across Snapdragon platforms. The announcement describes a glasses-platform demonstration and test results; it does not specify a release date for a glasses product.