Alibaba’s latest AI model, Qwen3.5, combines traditional large language model (LLM) capabilities with “advanced, dynamic reasoning.” This release marks a significant move in the competitive Chinese AI landscape, unveiling both an open-weight and a hosted version. The open-weight model allows users to download, fine-tune, and run it on their infrastructure, while the hosted version operates on Alibaba’s servers. With 397 billion parameters, Qwen3.5 supports native multimodal capabilities for understanding text, images, and video simultaneously, alongside new coding and agentic functionalities. It’s compatible with popular open-source AI agents and supports 201 languages, significantly increasing its reach from the previous 82. Benchmark tests claim performance on par with leading models from OpenAI, Anthropic, and Google DeepMind. Alibaba plans to release more open-weight models during the Chinese New Year, further solidifying its position against local rivals like ByteDance and Zhipu AI, who also updated their models recently.
Source link
