Alibaba Group Holding unveiled the preview version of its most powerful artificial intelligence model to date, solidifying its position as China’s leader in the race to catch American giants like Anthropic, Google and OpenAI.
Qwen3.5-Max-Preview, the flagship model of Alibaba’s Qwen 3.5 family, is now available on Arena, formerly known as LMArena, a model performance community created by researchers from UC Berkeley. Alibaba, which owns the South China Morning Post, unveiled the model on Thursday.
As of Friday, it was the top Chinese model on Arena’s list, but ranked 15th globally, behind products from Anthropic, which occupied the top two positions, and Google, whose Gemini-3.1-Pro-Preview came third.
But Qwen3.5-Max-Preview shows strong maths capabilities, ranking fifth globally in the category, trailing models including Anthropic’s Claude-Opus-4-6-Thinking and OpenAI’s GPT-5.4-High.
Alibaba said the new model was currently undergoing “final optimisations ahead of the release within the next two weeks”.
The preview is the latest evolution of the Qwen 3.5 series, which Alibaba has been aggressively expanding. Last month, the Hangzhou, Zhejiang province-based company announced several models in the family. These included Qwen-3.5-Open-Source, with 397 billion parameters, and Qwen-3.5-Plus, which offers one of the industry’s largest context windows at 1 million tokens – a reflection of how much data it can process at any given time.
