[Alibaba Qwen Releases Qwen3.8-Flash Model] Jinshi Data, August 26 – Alibaba Qwen has released the Qwen3.8-Flash model. According to the introduction, this is a multimodal MoE model and an early preview version of the Qwen4 architecture. The production version of Qwen3.8-Flash will soon be available via the Qwen Cloud API, priced at only $0.16 per 1 million input tokens and $0.47 per 1 million output tokens. The model features 125 billion parameters + 51 billion N-gram embedding parameters, but only activates 6 billion parameters per token, achieving extremely high cost efficiency.