ReleasesAug 27, 2026 · 18:434 sources

Alibaba Qwen Open Sources 6B Active Model as First Preview of Qwen4 Architecture

From huggingnews.com

AI;DR

Alibaba released the open-weight Qwen3.8-Flash-Next model, which uses a mixture-of-experts design with 6B active parameters per token. The model utilizes a hybrid attention mechanism and the Muon optimizer to reduce compute costs while reportedly outperforming DeepSeek-V4-Flash on coding and agent benchmarks. This release serves as a preview of the Qwen4 architecture for inference frameworks to adapt to before the formal launch.

Read the sourceSign in to suggest

Sources

+ 2 more
  1. SUPPORT@zhihufrontier“Flash-Next has 125B total parameters with only 6B active — far smaller than the 300B-class GLM-5.3-Flash”x.com
  2. SUPPORT@quixiai“running at 150 tok/s @ c1 to 661.1 tok/s @ c32 at 262k context”x.com

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today