Unverified60% confidenceFactExact time
Within the Qwen3 series there is a MoE model with a total parameter count of 2.4T and approximately 95B activated parameters, and downloading the model files requires approximately 5TB of storage space.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
8/19/2026
First Seen
Valid until: 11/17/2026
Sources
魔改2080Ti本地部署Qwen3 27B实测：FP8量化低成本方案
Related Claims
UnverifiedQwen 3.8 (27 billion parameters, Q4 quantization) requires reading approximately 17GB of memory per token75% similarVerifiedQLoRA quantizes the base model weights in 4-bit NF4 format, making it possible to fine-tune a 13-billion-parameter model on a single GPU.74% similarVerifiedQ4_K_M uses mixed precision, storing some critical weights at 6-bit and the remainder at 4-bit, with an overall average of approximately 4.5-bit.70% similarUnverifiedIn MoE architectures, activated parameters determine compute and speed, while total parameters determine memory, so 2B of activation does not make the 14.5 GB Q4_K_M file any smaller.70% similarUnverifiedMoE模型采用稀疏激活，Qwen 3.5 122B总参数122B但每次推理仅激活约10B，需将全部参数加载进内存，对内存容量极度敏感68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/775276API
curl https://kongchang.com/api/v1/knowledge/claims/775276MCP
get_claim(id=775276)