This generation carried the most concrete on-device generative claim Qualcomm has ever made for a mobile part — up to 10 billion parameters, up to 20 tokens per second, footnoted against Llama 2 7B. Both following generations dropped the parameter ceiling entirely, which makes 2023 the high-water mark for checkable Qualcomm AI figures.
Snapdragon 8 Gen 3
Qualcomm’s peak disclosure: an explicit parameter ceiling and token rate against a named model. Neither successor repeated it.