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
英伟达Nemotron开源MoE模型：笔记本即可本地运行的大语言模型
Related Claims
VerifiedQwen3-235B has a total parameter count of 235 billion, adopts a MoE architecture, and activates approximately 22B parameters during each forward inference pass.68% 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 adopts a Mixture of Experts (MoE) architecture with a total parameter count of 295 billion (295B), activating only 21 billion (21B) parameters per inference.63% similarUnverifiedMAI Thinking-E has a total parameter count of approximately 1T (trillion), with approximately 35B activated parameters, and uses a MoE (Mixture of Experts) architecture.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)