Qwen 3.8 may drop today as 27B dense model, killing MoE offloading speed
zyxciss · reddit · 2026-08-14
Reddit users speculate Qwen 3.8 could be released today as a 27B dense model. If so, it would abandon MoE sparsity, potentially making local inference much slower. Previously, Qwen 3.6 35B-A3B achieved 70 tok/s on an RTX 3060, but dense offloading could be 100x slower.
More from Models
- Tencent Releases UI-Mate-27B, a Desktop GUI Agent Model — tencent · 2026-08-24
- Sakana AI translation outperforms Google and DeepL in Japanese-English benchmarks — SakanaAILabs · 2026-08-24
- Developer haider makes his own LLM tier list after disagreeing with theo's rankings — haider1 · 2026-08-24
- Mystery OxAlpha Beats Claude; Alibaba Raises $10B for AI — 创业邦 · 2026-08-24
- OpenAI and Google cut LLM prices; mystery OxAlpha model beats Claude on DeepSWE — 创业邦 · 2026-08-24
- AI News Digest: DeepSeek Weekend Discounts, GPT-5.6 Sol Price Cut, Alibaba's $10B AI Raise — APPSO · 2026-08-24