Unverified50% confidenceFactExact time
xAI的Grok-1采用3140亿总参数、每次激活约25%专家的设计
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
VerifiedQwen3-235B总参数量为2350亿，采用MoE架构，每次前向推理实际激活的参数量约22B68% similarUnverifiedQwen3.6 35B模型包含256个专家，每个Token仅激活8个路由专家和1个共享专家，实际推理时激活参数约为3B64% similarUnverifiedHy3采用混合专家（MoE）架构，总参数量达2950亿（295B），每次推理仅激活210亿（21B）参数63% similarUnverifiedMAI Thinking-E总参数量约为1T（万亿），激活参数约35B，采用MoE（混合专家）架构63% similarUnverified苹果A18 Pro芯片的Neural Engine峰值算力达到约35 TOPS，足以本地运行30亿至70亿参数量级的压缩语言模型62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757424API
curl https://kongchang.com/api/v1/knowledge/claims/757424MCP
get_claim(id=757424)