Alibaba has unveiled Qwen3-Max, a new artificial intelligence model with more than one trillion parameters, as it intensifies its push into frontier AI technology.
The model was announced at the company’s annual Apsara conference, where Alibaba executives also introduced Qwen3-Omni, a multimodal system designed for immersive and extended-reality applications. Qwen3-Max, the company said, demonstrates particular strength in code generation and autonomous agent tasks, allowing it to perform multi-step operations with fewer prompts.
Alibaba positioned the model as a rival to the latest systems from Western and Chinese competitors. Benchmark results shared by the company cited Tau2-Bench, a reasoning test, and showed Qwen3-Max outperforming Anthropic’s Claude and DeepSeek’s V3.1 models on selected metrics. Independent verification of those results has not yet been released.
“Qwen3-Max represents a significant leap forward in our mission to integrate advanced AI into practical business applications,” Alibaba Cloud Intelligence chief executive Jingren Zhou said at the launch. “By expanding both the scale and the reasoning capabilities of our models, we are enabling customers to unlock entirely new opportunities across industries.”
The launch comes as Alibaba continues to position artificial intelligence at the centre of its long-term strategy. In March, chief executive Eddie Wu pledged to increase the group’s three-year, 380 billion yuan (£41 billion) investment in AI infrastructure. Alibaba’s New York-listed shares rose on Tuesday following the Qwen3-Max announcement.
Qwen3-Max is the largest release so far in Alibaba’s Qwen family, which the company began scaling in April with both dense and sparse model variants. The trillion-parameter threshold places Alibaba among a small group of technology companies worldwide seeking to train models of this magnitude.
Analysts said the launch highlights China’s determination to keep pace with U.S. rivals, despite challenges from export controls on high-end chips. Trillion-parameter models, often using mixture-of-experts designs, demand enormous computing resources to train and deploy. Alibaba did not disclose details of the architecture or hardware underpinning Qwen3-Max, nor whether the model will be made openly available beyond Alibaba Cloud customers.
As competition among large model makers accelerates, the company’s focus on autonomous agent tasks and enterprise integration suggests a strategy aimed at embedding Qwen systems across its cloud and commerce ecosystem. Whether the model can deliver efficiency and reliability at scale will determine its impact in a market already crowded with claims of technical leadership.
