Unverified50% confidenceFactExact time
在MoE模型中每个token只激活少量Expert（通常2-4个，Top-K路由），实际计算量可能仅相当于6-7B的Dense模型
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedMoE模型中'35B'指总参数量，每个token只激活少量Expert（通常2-4个），实际计算量可能仅相当于6-7B的Dense模型84% similarUnverifiedMoE models activate only a small fraction of total parameters per inference; Qwen3 Max may activate between 1/8 and 1/16 of the total parameters per inference.74% similarUnverifiedKimi K2.7 Code uses a MoE architecture with 384 experts, selecting 8 experts plus 1 shared expert per token, with actual computation approximately 2.3% of that of a dense model.70% similarUnverified千问3.6的35B MOE模型每次只激活其中3B的参数70% similarVerifiedMixtral 8x7B模型拥有8个专家，每个Token只激活2个，总参数量达到47B，但单次推理的计算量仅相当于一个13B参数的稠密模型68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/575452API
curl https://kongchang.com/api/v1/knowledge/claims/575452MCP
get_claim(id=575452)