Expired50% confidenceFactExact time
The Qwen3.6 34B 80-layer model was created by community developers who expanded the 64-layer Transformer network of Qwen3.6 27B to 80 layers, growing the parameter count from 27B to 34B, an increase of approximately 26%
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Related Claims
UnverifiedQwen3.6 27B supports a context length of 262K, and the 34B 80-layer version maintains the same context length unchanged77% similarUnverifiedThe 40B OPUS distilled version expands the Qwen3.6 27B model to 40B parameters.72% similarUnverifiedQwen3-235B-A22B adopts a MoE architecture, activating only 22B out of 235B total parameters at a time.70% similarUnverifiedQwen3-series coding models load at approximately 20 GB and can expand to 35–40 GB during inference.69% similarUnverifiedWhen deploying the Qwen3 27B NVFP4 model on a 5090 (32 GB VRAM), vLLM can support a context length of approximately 70K.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40887API
curl https://kongchang.com/api/v1/knowledge/claims/40887MCP
get_claim(id=40887)