Unverified50% confidenceFactExact time
Qualcomm partnered with Moonshot AI to run a 30-billion-parameter model on the Snapdragon 8 on-device platform, completing memory adaptation in collaboration with Wulianghuo and Jiangbo Long; memory requirements are reduced by more than 50% compared to conventional solutions, prefill throughput exceeds 330 tokens/s, and it can run fully offline.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/30/2026
First Seen
Valid until: 12/29/2026
Sources
Related Entities
Related Claims
UnverifiedWithout the 3090s connected, the machine can serve as a low-power server, running small models using only the 7840U, 64GB of memory, the Vulkan backend, and llama.cpp.66% similarUnverifiedIf all 36 layers of Spark X 2.5 tracked the full text, the KV cache overhead for tracking 1 million tokens would be approximately 155 GB, while the sparse design compresses this to approximately 39 GB66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961391API
curl https://kongchang.com/api/v1/knowledge/claims/961391MCP
get_claim(id=961391)