https://modelscope.cn/models/Qwen/Qwen3.8-Flash-Next/summary
Qwen3.8-Flash-Next is a multi-modal MoE model built on the next-generation Qwen4 architecture. By introducing new architectural changes such as the GDN hybrid architecture and the Qwen Sparse Attention mechanism (QSA), Qwen3.8-Flash-Next explores new horizons for the capabilities and performance of next-generation models. It demonstrates outstanding abilities in areas such as agent coding, long-horizon tasks, and multi-modal intelligence. To prepare the community for the upcoming release of the complete Qwen4 model family, these architectural improvements are being made public in advance.
