Sources revealed that Alibaba will open-source its next-generation large language model, Qwen3.5, tonight on Chinese New Year's Eve. The model features comprehensive architectural innovations, adopting a hybrid attention mechanism that enhances long-text processing capabilities by over 40%. It supports native multimodal visual understanding and plans to open-source both a 2B dense model and dual versions of 35B-A3B MoE to lower deployment barriers for developers.
