Issue 2026-11-08 · Industry · 模型发布 · 开源
Tencent releases Qwen3.8-27B-DSpine-b16 model
A Hugging Face listing shows Tencent's Qwen3.8-27B-DSpine-b16. Community posts say it uses a new speculative decoding method that may speed up decoding for 27B models by about 3x; the claims come from a Reddit post and await official confirmation.
Read original ↗- Community questions open-source models dropping security capabilities
- Tencent releases Hy-Memory-30B-A3B model
- Tencent releases 5B omni-modal parsing model Youtu-Parsing-Omni
- Opinion: AI Has Crossed a Line and Should Be Regulated Like Banks
- Open-source inside-llm visualizes LLM internals and quantization gaps