Unverified60% confidenceFactExact time
Kimi K2.7 Code采用384个专家的MoE架构，每个token选择8个专家加1个共享专家，实际计算量约为稠密模型的2.3%
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
Kimi K2.7 Code Deep Dive: The Cost-Performance Battle of a Trillion-Parameter Coding Model
bilibiliAI睡前简报6/13/2026
Related Claims
VerifiedKimi K2.6采用混合专家（MoE）架构，总参数量达1万亿级别，实际激活参数仅32B74% similarUnverified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70% similarUnverifiedIn the Gemma 4 26B MOE model, the model contains 128 internal 'experts' but only 8 are activated per inference.67% similarUnverifiedKimi K2.7 Code共有61层，包含1个稠密层67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/41723API
curl https://kongchang.com/api/v1/knowledge/claims/41723MCP
get_claim(id=41723)