Verified65% confidenceFactExact time
35B-A3B is a MoE architecture naming convention, where 35B refers to a total parameter count of approximately 35 billion, and the A in A3B stands for activated parameters of approximately 3 billion.
3
Sources
65%
Confidence
Long-term
Relevance
8/19/2026
First Seen
Sources
Qwen开发者暗示别等35B-A3B：社区三大猜测解读
redditr/LocalLLaMA8/18/2026
Related Claims
VerifiedQwen3.6 35B-A3B是混合专家架构（MoE），总参数量35B，每次推理实际激活参数量约3B78% similarUnverifiedgemma-4-26b-a4b 模型名称中的 'a4b' 表示该模型采用了混合专家（MoE）架构，激活参数量约为4B74% similarUnverifiedThe 35B A3B version ranked first in multiple evaluations with balanced capabilities across various domains73% similarVerifiedHy3 adopts a Mixture of Experts (MoE) architecture with a total parameter count of 295 billion (295B), activating only 21 billion (21B) parameters per inference.72% similarUnverifiedOrnith 1.5 35B-A3B uses a MoE architecture with a total parameter count of approximately 35B, activating only approximately 3B parameters per inference.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/775685API
curl https://kongchang.com/api/v1/knowledge/claims/775685MCP
get_claim(id=775685)