Qwen3.8-Flash-Next on B.AI

Overview

Alibaba’s Tongyi Qianwen team released Qwen3.8-Flash-Next, a highly sparse MoE model with 125B total parameters but only 6B active per token, integrating hybrid attention (GDN+QSA) and extending context up to 1M tokens via YaRN.[3] On B.AI, the model is offered as a forward-looking preview of Qwen4 with free API access for programming assistance, agent workflows, and document analysis, making it a strong choice for developers needing long-context coding and data reasoning from China.[3][7]