Verified65% confidenceFactExact time
The modern form of MoE can be traced back to a Google paper in 2017, and in 2022 Google's Switch Transformer demonstrated that trillion-parameter sparse models can be trained efficiently.
3
Sources
65%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
Tencent Hunyuan HY3 Official Version Open-Sourced: The Strongest Under 300B, at 1 Yuan per Million Tokens
bilibiliinfinite灵感港7/6/2026
Related Claims
VerifiedThe MoE architecture was proposed by Shazeer et al. in 2017 in the paper "Outrageously Large Neural Networks" and subsequently validated at scale by Google in Switch Transformer (2021)82% similarUnverifiedGoogle于2021年发布的Switch Transformer论文首次证明MoE可将训练效率提升4倍以上81% similarUnverifiedMoE混合专家架构最早可追溯至1991年的学术论文，谷歌于2021年发布Switch Transformer将专家数量扩展至2048个80% similarVerifiedMoE was reactivated in the era of large language models by Google's Switch Transformer (2021) and GLaM.80% similarUnverifiedGoogle于2022年发布的Switch Transformer证明MoE架构在相同计算预算下收敛速度可达稠密模型的4-7倍79% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/486081API
curl https://kongchang.com/api/v1/knowledge/claims/486081MCP
get_claim(id=486081)