Aliyun unveils new Qwen3.8-Flash-Next model based on next-gen qwen4 architecture

Aliyun has announced the upcoming release of the Qwen3.8-Flash-Next and Qwen3.8-Flash-Next-FP8 models, set to be open-sourced at 11:00 PM Beijing Time on August 26. The Qwen3.8-Flash-Next model is based on the next-generation Qwen4 architecture and is a multimodal Mixture of Experts (MoE) model. According to official statements, these early releases of the Qwen4 architecture aim to prepare the community for the upcoming full Qwen4 model family. The models will be available on the ModelScope platform. Additionally, Aliyun has previously released the Qwen3.8-27B model, which outperforms the Qwen3.7-Plus in programming and office scenarios. The Qwen3.8-2.4T-A95B model, with 2.4T MoE, activated 95B parameters, and native 256K context length, was also made available to the public. The release of these models highlights Aliyun's ongoing efforts to advance large language model capabilities and expand their accessibility to the broader community.