Qwen3.8-Flash-Next is an open-weight multimodal MoE model (125B total + 51B N-gram embeddings, 6B active) released as an early architecture preview for the upcoming Qwen4 family. It introduces hybrid Gated DeltaNet + QSA attention, gated residuals, and Muon optimizer, delivering stronger coding and office-task performance than Qwen3.7-Plus at roughly 1/9th the training cost, with native 262K context (extensible to 1M) and support for text, image, and video inputs.

Read the official Qwen (Alibaba) launch page

Worth opening first

Standouts

Playable

Locked Protocol v1-show

A side-by-side Three.js evaluation and gallery framework for AI models.

Playable1

Playable

Locked Protocol v1-show

A side-by-side Three.js evaluation and gallery framework for AI models.

Apps

Built something with Qwen3.8-Flash-Next?

Paste the URL and Dropday.ai will credit you, track the build as it evolves, and put it on this drop.

Submit a build