Your Next Phone Could Take Notes and Talk Back Without the Cloud
Qualcomm announced two new flagship phone chips, Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, at its annual Snapdragon Summit. The main goal is more personal AI assistants on the phone. TechCrunch
Both chips use a low-power sensing hub that runs small AI models of up to 200 million parameters. Parameters are a measure of model size and ability. Like a helper that stays awake all day, the hub does steady background work without sending each task to the main processor.
The hub lets a phone run its own note-taker and tell speakers apart. Transcripts and speaker labels stay on the phone. It also remembers phone use to suggest task automations. Qualcomm said the new chip can run a full agent that listens and speaks.
Listening, using on-phone context, and speaking happen in one on-device loop. That cuts routine network trips and keeps personal voice data close to the user.
The two chips offer different amounts of AI power. The regular Gen 6 has a new accelerator part to run models more efficiently. Qualcomm did not bill it as raw speed. The point is performance per watt, or more work per unit of battery.
The Extreme can run a 30-billion-parameter mixture-of-experts model on the device. That design splits the model into expert parts that switch on as needed. Apple released a 20-billion-parameter model of that type at its Worldwide Developer Conference in June. For Android developers, the 30-billion limit sets what can stay in memory without cloud help.
Camera and audio improve with AI. The Extreme records 8K video at 60fps and 4K at 240fps for slow motion. It supports the Advanced Professional Video (APV) codec for pro recording. Both chips use AI to boost voices and cut noise.
Both include voice bubble tech for calls, separate from basic noise reduction. It keeps a quiet zone around the caller so speech stays clear in crowds.
Motorola announced the Motorola Signature 27 with the Extreme chip. It is an early flagship outlet and a test target for bigger on-device models and the new video tools.
The broader context here is the two-level plan. The small hub tracks routines and speakers all day. The big model handles hard questions only when asked.
In my view, that split makes sense. An assistant is several models with different needs for speed, power, and memory. If note-taking, speaker sorting, memory, and voice replies live on the phone, apps will stand out less for cloud access and more for how they use permissions, sensors, and app actions. Over time, always-on small models plus big models on standby could make phones more helpful without sending each request across the network.
