Unverified50% confidenceFactExact time
The MoE architecture was first systematically introduced into the Transformer framework by Google and was subsequently advanced by leading Chinese model companies.
1
Sources
50%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
VerifiedThe MoE architecture was first proposed by Jacobs et al. in 1991 and in recent years has been brought back into the mainstream by models such as Google's Switch Transformer and Mixtral.78% similarVerifiedAfter the successful deployment of MoE models represented by Google Switch Transformer and Mixtral, the MoE architecture has become the mainstream technical choice for ultra-large-scale language models.77% similarVerifiedThe MoE architecture concept can be traced back to the 1990s, was reignited in the large model era by Google's Switch Transformer, and has been validated by projects such as Mixtral and GPT-4.76% similarVerifiedThe 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)75% similarUnverifiedThe 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.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/590224API
curl https://kongchang.com/api/v1/knowledge/claims/590224MCP
get_claim(id=590224)