- Published on
Qualcomm Unveils Two Sixth-Generation Snapdragon 8 Elite Flagship Chips, Focusing on On-Device AI and Agent Capabilities
- Authors
- Name
- DeepHorus
Qualcomm officially unveiled two new generations of flagship mobile processors at its annual Snapdragon Summit on September 22 local time β the Snapdragon 8 Elite Gen 6 and the Snapdragon 8 Elite Extreme Gen 6. Both chips place AI at the core of their upgrade direction, with a key focus on strengthening smartphones' capabilities for personalization when running AI agents, on-device inference, and multimodal processing, marking a shift in the design focus of flagship mobile processors toward AI computing.
In terms of specific capabilities, the new-generation chips are equipped with a new sensing hub that can run lightweight AI models with up to 20 billion parameters locally. This enables smartphones to directly run personal journaling assistants on-device, distinguish between different speakers' voices, and build longer-term "memories" based on the user's daily usage patterns, thereby providing more accurate suggestions for automated tasks. The new sensing hub also supports fully capable voice-input and voice-output agents, from listening to commands and understanding semantics to generating voice replies, all of which can be completed locally, reducing dependence on cloud connectivity and network connections.
At the AI computing architecture level, the Snapdragon 8 Elite Gen 6 introduces a new accelerator unit optimized for model architectures commonly seen in generative AI and agent applications, rather than simply piling on larger-scale computing resources. The higher-end Extreme variant can run mixture-of-experts (MoE) models with up to 30 billion parameters locally on the phone, activating only a subset of parameters when handling specific tasks, thereby maintaining a larger model scale while reducing the computational load required for inference.
For comparison, Apple announced a mixture-of-experts model with 20 billion parameters at its Worldwide Developers Conference this June, described as the most advanced model among its third-generation foundation models. Qualcomm's move to bring a 30-billion-parameter-level MoE model to mobile devices further narrows the gap between flagship smartphones and large-scale generative AI models. This capability also means that the competitive focus of on-device AI is shifting from "whether a model can run" to "how large a model can run."
Beyond AI, the new-generation chips also enhance imaging and audio capabilities. The new CPU provides more granular pixel-level camera control and improves video stabilization and motion understanding capabilities; the Extreme variant supports up to 8K 60fps video recording and 4K 240fps ultra-high-definition capture, and adds Qualcomm's new advanced professional video encoding format APV, targeting professional-level shooting and post-production needs. Both chips can also use AI to improve voice quality, reduce ambient noise, and support "Voice Bubble" technology, which separates the user's voice from surrounding noise during calls.
Meanwhile, Motorola announced the launch of the Motorola Signature 27 phone equipped with the Snapdragon 8 Elite Extreme Gen 6, expected to officially hit the market later this year, meaning the new chip will soon enter the consumer market. Qualcomm previously revealed that it is working with partners to develop more than 40 AI devices, covering new form factors such as smart glasses and wearables, but the company still believes that smartphones will continue to be an important device for consumers to use AI, as they already have mature cameras, microphones, displays, communication capabilities, and a wealth of user data.
Qualcomm's product roadmap this time reflects the trend that mobile AI is shifting from simple generative features to a more continuously running "agent" model. In the past, mobile AI was mainly used for standalone tasks such as photo processing, text generation, translation, and voice assistants, whereas the new platform aims to enable AI to continuously perceive user behavior and proactively offer suggestions or even execute tasks based on context. This is also the reason why Qualcomm specifically emphasizes the sensing hub and local memory capabilities β agents not only require powerful models but also need to continuously access environmental and historical information, and deploying them on the chip can reduce data uploads to the cloud while lowering latency.
As the new-generation chips gradually enter the market, competition among flagship smartphones is likely to increasingly revolve around on-device AI. Whether phones can run larger models, understand more complete personal context, complete complex tasks without relying on the cloud, and run continuously under manageable power consumption will all become important indicators in the next phase of mobile platform competition.
