Qualcomm's New Flagship Chips Move Personal AI Agents Onto the Phone
Qualcomm announced Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6 at its annual Snapdragon Summit, with improved personalization for AI agents as the central task for its next flagship tier. TechCrunch
The design centers on sensing hubs able to run small models of up to 200 million parameters, the values that set what a model can do. Those hubs handle persistent, low-power contextual work without sending every interaction to the main processor.
The hub lets a phone run a personal scribe locally and separate speakers, so transcription and speaker labels stay on device. It can also build memory from usage to suggest task automations. Qualcomm said the new chip can run a complete voice-in and voice-out agent.
That setup brings speech recognition, reasoning over local context, and speech synthesis into one on-device loop. It lowers reliance on network round trips for routine agent work and keeps personal audio and usage-derived memory close to the user.
The two chips differ in AI headroom, or how much AI work they can sustain at once. The 8 Elite Gen 6 adds a new accelerator element built to run models more efficiently. Qualcomm did not present it as a raw speed gain. The emphasis is performance per watt, or sustained agent and sensing work for less energy.
The Extreme version can run a 30-billion-parameter mixture-of-experts model locally. In such a model, smaller expert sections switch on as needed for a query. Apple released a 20-billion-parameter mixture-of-experts model at its Worldwide Developer Conference in June. For flagship Android developers, the 30-billion figure sets what can stay resident in memory and run without cloud fallback.
Imaging and audio get parallel upgrades tied to AI processing. The Extreme supports 8K video at 60fps and 4K at 240fps for slow motion, plus the Advanced Professional Video (APV) codec for pro-level recording. Both chips use AI to lift vocals and cut noise.
Both chips include Qualcomm's voice bubble technology for calls, distinct from general noise reduction. It forms an isolation zone around the user, keeping speech intelligible in crowded acoustic settings.
Motorola announced the Motorola Signature 27 powered by the Snapdragon 8 Elite Extreme Gen 6. That provides an immediate flagship outlet for the Extreme part and a concrete target for testing larger local models and the new video pipeline.
The broader context here is the split between a 200-million-parameter sensing tier and a 30-billion-parameter expert tier. The hub tracks routines and holds speaker-aware local state over time. The large model steps in for complex prompts when invoked.
In my view, that division is pragmatic. Agents are not one model but a system of models with different budgets for delay, power, and memory. For software teams, the effect is a move in where differentiation sits. If transcription, speaker separation, usage memory, and voice synthesis run locally by default, the baseline agent depends less on cloud model access and more on links to on-device permissions, sensors, and app actions. Over time, small models running continuously with larger models on call could make phones more capable assistants without turning every request into a network transaction.
