Unverified85% confidenceFactTime unknown
Qwen3.5的参数激活率为4.3%，而DeepSeek V3的671B模型激活率为5.5%
1
Sources
85%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Qwen3.5深度解析：混合注意力架构实现19倍长上下文加速
bilibili雪天鱼
Related Entities
Related Claims
UnverifiedQwen3-4B在多项推理类评测指标上已与DeepSeek V3或GPT-4o相差无几79% similarUnverifiedQwen3旗舰版（235B）的参数量仅为DeepSeek旗舰版（671B）的三分之一左右77% similarPartially VerifiedQwen 3 Max和DeepSeek-V3均采用MoE架构，总参数量达数千亿，但实际推理激活参数仅为总量的一小部分75% similarUnverifiedQwen 3.6 Max在中性模式下端到端准确率仅12.3%75% similarUnverifiedQwen3.6 27B Q4量化版总分650分，排名第四74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/18078API
curl https://kongchang.com/api/v1/knowledge/claims/18078MCP
get_claim(id=18078)