Alibaba to Open Source Qwen3.8-Flash-Next Multimodal Model
Alibaba will release the source code for its Qwen3.8-Flash-Next multimodal MoE model on August 26, built on the next-generation Qwen4 architecture. The release aims to prepare the community for the upcoming Qwen4 model family.
Alibaba is set to open source its Qwen3.8-Flash-Next multimodal Mixture-of-Experts (MoE) model on August 26, according to announcements on the ModelScope community platform. The model is built upon the next-generation Qwen4 architecture.
Two versions of the model, Qwen3.8-Flash-Next and Qwen3.8-Flash-Next-FP8, will be made available. Alibaba stated that releasing these architectural improvements in advance is intended to allow the developer community to prepare for the upcoming full Qwen4 family of models.
Further details regarding the specific capabilities and performance of these models have not yet been disclosed. The release follows Alibaba's recent open-sourcing of other Qwen models, including Qwen3.8-27B, which demonstrated improved performance in programming and office tasks compared to its predecessor.
In addition to Qwen3.8-27B, Alibaba has also previously released the Qwen3.8-2.4T-A95B model, notable for its large scale and context window.