PrismML puts a 2B model on Snapdragon glasses

PrismML demonstrated a 2-billion-parameter 1-bit vision-language model running locally on Qualcomm Snapdragon AR1 Gen 1 smart glasses, with Qualcomm testing 15.36 tokens per second and a 1,024-token context window in a PrismML announcement. TechCrunch’s report says the 1-bit model uses roughly 4× less memory than a comparable 4-bit model; edge-AI teams should treat quantization and model–hardware co-design as a product lever, not a benchmark footnote, as local inference economics shift toward capability per watt.