Cailian Press reports that the code for Alibaba’s latest large language model, Qwen3.5, has been submitted for a merge request on the open-source platform HuggingFace—a clear indicator that its official open-sourcing is imminent. Qwen3.5 introduces an innovative hybrid attention mechanism and is likely to be a vision-language model (VLM) with built-in support for visual comprehension. Anticipated open-source releases include at least a 2-billion-parameter dense model and a 35-billion-to-3-billion-parameter Mixture of Experts (MoE) model.
