Chinese tech giant Alibaba unveiled Qwen-3 Max, its largest-ever AI model, at the company's 2025 Apsara Conference in its home city Hangzhou, east China's Zhejiang Province, on Wednesday.
The Qwen-3 Max model has over 1 trillion parameters and was pretrained on 36 trillion tokens It comes with a context length of one million tokens, which allows it to reason through entire codebases or lengthy documents.
The new model is said to offer improvements in reasoning, instruction following, multilingual support, and long-tail knowledge coverage. A preview version of Qwen3-Max released earlier this month was ranked No.3 on LMArena's text leaderboard.
"Compared to 2022, the year of generative AI, the scale of energy consumption by the Alibaba Cloud's global data centers is projected to increase tenfold by 2032. This is part of our long-term plan, as we believe that such extensive investment will drive the development of the AI industry and prepare us for the arrival of the ASI (Super Artificial Intelligence) era," Wu Yongming, chief executive officer of Alibaba Group, told the audience at the event.
The three-day annual event will showcase Alibaba's latest advancements in infrastructure, platforms, AI models and applications. The company unveiled seven large model products, covering various fields including language, speech, vision, multimodal applications, and code models.
Alibaba unveils largest-ever Qwen-3 Max model
