Unverified85% confidenceFactTime unknown
40B OPUS蒸馏版将Qwen3.6 27B模型扩展到了40B参数量
1
Sources
85%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
Qwen3.6 27B三大邪修量化模型实测：代码暴增15.8PP、40B蒸馏、16GB适配
bilibili尚书省说书人
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77% similarUnverifiedQwen3-235B-A22B采用MoE架构，235B总参数中每次只激活22B76% similarUnverifiedQwen3系列编程模型约20GB加载，推理时可膨胀到35-40GB75% 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%72% similarUnverified一块24GB显存的RTX 4090可流畅运行Qwen3-32B的4-bit量化版本71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/18016API
curl https://kongchang.com/api/v1/knowledge/claims/18016MCP
get_claim(id=18016)