Unverified90% confidenceFactTime unknown
Qwen3-235B-A22B采用MoE架构，235B总参数中每次只激活22B
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Unsloth：本地微调大模型速度提升5倍的开源神器
githubunslothai
Related Entities
Related Claims
UnverifiedQwen3.6 27B支持262K上下文长度，34B 80层版本保持相同的上下文长度不变79% similarUnverified40B OPUS蒸馏版将Qwen3.6 27B模型扩展到了40B参数量76% similarUnverified千问3（Qwen3）旗舰版参数量高达235B、体积约142GB71% similarUnverifiedQwen3.6 34B 80层模型是由社区开发者将Qwen3.6 27B的64层Transformer网络扩展到80层，参数量从27B增长到34B，增幅约26%70% similarUnverifiedMuse-Glimmer-30B参数规模为30B级别（实测架构约28B），支持128K上下文，并在发布时提供本地量化版本70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33220API
curl https://kongchang.com/api/v1/knowledge/claims/33220MCP
get_claim(id=33220)