Qualcomm announced Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6 on September 22, expanding its flagship smartphone processor lineup with hardware aimed at running personalized AI agents locally.

The Extreme model's product brief lists a new Element Accelerator within the Hexagon neural processing unit, alongside scalar, vector and tensor processing components. Qualcomm also specifies 50% more shared NPU memory and support for numerical formats ranging from INT2 to FP16, including mixed precision.

The company claims 35% higher NPU performance and 33% better AI performance per watt than Snapdragon 8 Elite Gen 5. Those are vendor comparisons, not independent handset benchmarks; Qualcomm explicitly says results depend on device-maker implementation and other factors.

Sensing hardware for local context

The Extreme platform's Sensing Hub includes dual micro NPUs for audio, voice and sensor processing to create personal knowledge graphs. Its specifications also list Qualcomm Personal Scribe. Qualcomm describes combining the sensing hub, NPU and Oryon CPU to learn routines, preferences and environmental context directly on the device.

TechCrunch separately reports Qualcomm's claim that the Extreme chip can run a 30-billion-parameter mixture-of-experts model locally. That parameter count is not specified in the product brief, and does not by itself establish response speed, memory requirements or battery consumption in a shipping phone.

For developers, the announcement points to more dedicated hardware for local inference and contextual processing. It does not establish that every handset using these chips will expose the same agent features or deliver the same sustained performance.