Change language to
0:00

Qualcomm says its upcoming Snapdragon 8 Elite Gen 6 can run Mixture-of-Experts AI models with up to 30 billion parameters on a phone, although the chip names are not yet official and the claim is about model size rather than 30 billion parameters being active at once.

Android Authority reports that Qualcomm’s next-generation Hexagon NPU is designed for on-device multimodal AI and agentic workflows. The company says its new Element Accelerator is built for transformer inference, while a larger shared-memory pool keeps more model state and KV-cache data closer to the processing units.

Qualcomm Hexagon NPU diagram showing 50 percent more shared memory

Snapdragon 8 Elite Gen 6 and 30B AI models

The 30-billion-parameter figure comes from a Mixture-of-Experts architecture. Instead of activating every parameter for each response, the model routes each token through a smaller group of specialised experts. Qualcomm’s example activates about 3 billion parameters at a time, while the larger model remains available on the device.

HotHardware reports that the new NPU has 50 per cent more shared memory than the Snapdragon 8 Elite Gen 5 and supports context lengths of up to 32,000 tokens. Qualcomm also claims faster pre-fill and decoding for INT4 models, but those figures describe specific workloads rather than a universal 50 per cent increase in AI speed.

Qualcomm Hexagon NPU slide showing support for 30B MoE models

What changes for phones in the UAE?

Qualcomm has not announced phone models, UAE prices or local release dates for the new platform. The company is expected to reveal the next Snapdragon flagship chips at Snapdragon Summit, which starts on September 22. Phones using the chips should follow from manufacturers, but their UAE availability will depend on each brand’s local launch plans.

The on-device approach could reduce how often a phone sends prompts to the cloud, particularly for agentic tasks that need to keep context and call tools. The actual experience will depend on memory, software support, thermals and the model a phone maker ships. Qualcomm has described the capability; it has not shown a retail phone running a 30B model.

Qualcomm Hexagon NPU Element Accelerator with 32K context support

Qualcomm’s previous [Snapdragon 8 Elite Gen 6 5GHz CPU](https://tbreak.com/snapdragon-8-elite-gen-6-5ghz/) and [Adreno Neural Fusion GPU](https://tbreak.com/qualcomm-adreno-neural-fusion-gpu-next-snapdragon/) announcements have covered the platform’s CPU and graphics upgrades. The NPU is the third part of the platform preview, with the full specifications still to come at Snapdragon Summit.

NEWSLETTERS

Subscribe to our Newsletters

Two newsletters. Zero noise. Pick what lands in your inbox.

Unsubscribe anytime. We don’t share your email.

Can the Snapdragon 8 Elite Gen 6 run 30B AI models?

Qualcomm says its new Hexagon NPU can run Mixture-of-Experts models with up to 30 billion parameters locally, activating only a smaller group of parameters for each token.

When will Qualcomm reveal the Snapdragon 8 Elite Gen 6?

Qualcomm is expected to reveal the next Snapdragon flagship chips at Snapdragon Summit, which starts on September 22.

Will Snapdragon 8 Elite Gen 6 phones launch in the UAE?

UAE phone models, prices and release dates have not been announced. Local availability will depend on individual manufacturers.