Unverified50% confidenceFactExact time
The three operators Mgroup GEMM, MQA Logits, and MGroup MoE are deeply tied to the specific architectures of new models such as DeepSeek V4.1 Flash.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/3/2026
First Seen
Valid until: 1/1/2027
Sources
Related Entities
Related Claims
UnverifiedDeepSeek-V4 Flash采用混合注意力、mHC混合机制以及哈希路由的MoE等架构设计73% similarUnverifiedDeepSeek-V4.1-Flash adopts a MoE (Mixture of Experts) backbone architecture with 552B parameters.70% similarUnverifiedThe models actually run on this system include open-source large language models such as GLM, the DeepSeek Flash series, the Qwen3 series, KimiK, and MiniMax.67% similarUnverifiedv0.5.21 adds support for LLM/VLM models including DeepSeek-V4.1 Flash, GigaChat 3.5, IQuest-Q1, the MiMo-V2.6 series, and Ling-3.0-flash-VL.67% similarUnverifiedRecently popular models such as DeepSeek-V2/V3 and Qwen2-MoE all adopt the MoE architecture.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/970645API
curl https://kongchang.com/api/v1/knowledge/claims/970645MCP
get_claim(id=970645)