Qwen3.8-Flash-Next

Alibaba's Qwen team released Qwen3.8-Flash-Next, an open-weight multimodal MoE model previewing the Qwen4 architecture. It has 125B total parameters with 6B active, a 51B N-gram embedding layer, and a 262,144-token context window. It outperforms Qwen3.7-Plus at roughly one-ninth the training cost.