Qualcomm has adapted the StepEdge-Omni 30B-MoE model to run on smartphones. The MoE technique reduces memory demand by more than 50%. The model allows execution without using the cloud.
Prefill speed is 330 tokens/s and decoding speed is 28 tokens/s. Qualcomm highlighted that the model can perform tasks such as understanding emails, planning trips and suggesting accommodations. The partnership with StepEdge, Wulanghuo and Jiangbolong was announced to adapt the model to the Snapdragon 8 Ultra. Huawei also unveiled new AI technologies in Shanghai, such as the SuperPoD architecture and UnifiedBus technology.
The partnership includes adaptations for the Snapdragon 8 Ultra, which supports running the model on smartphones. Huawei, for its part, presented its SuperPoD architecture and UnifiedBus technology in Shanghai, which connect up to 4,000 processors with fast and efficient optical connections. These innovations reflect the race for efficiency in artificial intelligence, with companies such as Darktrace offering security solutions to monitor corporate use of the technology.
Speed and efficiency in text processing
Decoding speed also helps answer questions with less delay. The memory reduction allows the model to work on devices with less capacity. The partnership with Jieyu, Wulanghuo and Jiangbolong made it possible to tune the model for the Snapdragon 8 Ultra, ensuring performance on mobile devices. This adaptation opens the way for new applications that don't depend on the cloud.
Qualcomm is developing AI models for mobile devices in partnership with companies such as StepEdge, Wulanghuo and Jiangbolong. Huawei is also launching new AI technologies to reduce dependence on Nvidia components. The partnership between Qualcomm and StepEdge shows how the industry is moving to integrate AI into mobile devices efficiently.
The partnership between Qualcomm and StepEdge is part of a larger trend of integrating AI into mobile devices, with companies such as Wulanghuo and Jiangbolong also involved. The expansion of AI to mobile devices is accelerating, with security solutions such as Darktrace's SECURE AI helping to monitor corporate use of this technology.
AI helps predict unexpected threats
McCall McIntyre of OpenAI said that working with partners like Darktrace is "critical to turning advances in AI into stronger, more practical defenses" (loosely translated).
The collaboration helps translate innovations into real tools against threats. Darktrace is developing capabilities under the SECURE AI program, available since February. Qualcomm is adapting a 300-billion-parameter AI model to run on smartphones without the cloud, using an MoE technique that cuts memory demand by more than 50%. The technology could change how AI is used on mobile devices, without depending on the internet.
The partnership with StepEdge, Wulanghuo and Jiangbolong speeds up this evolution. This technology reduces energy consumption, allowing the model to operate on low-power chips.
Still unconfirmed
- Details about the AI model, its features and costs were not disclosed
- There is no information about the HighTech announcement in primary or specialized sources
With information from phone.cnmo.com, ithome.com and scmp.com.
Sources
- phone.cnmo.com
- ithome.com
- Huawei unveils latest tech boost ai power push break chinas nvidia reliancescmp.com
- Exclusive darktrace makes secure ai generally available to monitor enterprise ai usesiliconangle.com
- medium.com
- Codenbuild technologies building the future with ai software practical technology educationmedium.com
Text written with AI from the sources listed and checked by the newsroom's automated rules. How we work
Want to see resumer first in Google searches?
Add to Google's preferred sources


