According to CoinMeta, Alibaba announced that it will release a multimodal MOE (hybrid expert model) based on the Qwen4 architecture on August 26th at 23:00. The official statement indicates that releasing this architecture in advance is to prepare the community for the subsequent complete Qwen4 model family. This release is not the official version of Qwen4, but rather a technical preview. At the beginning of this month, Alibaba released a Qwen3.8-Max with 2.4T parameters, which is quite fast pace.