Alibaba’s Qwen3-Max Joins the Frontier of Trillion-Parameter AI Models September 24, 2025 by Jaime Hampton 
(QINQIE99/Shutterstock)
Alibaba has introduced its most advanced AI model yet with the launch of Qwen3-Max, the latest in the company’s Qwen series of models. The company says Qwen3-Max surpasses competitors in coding, reasoning, and autonomous task performance. The release is in line with Alibaba’s push to become a central player in the global large-scale AI race and comes with a pledge from chief executive Eddie Wu to accelerate spending on AI and cloud infrastructure. The news helped send Alibaba shares to their highest level today since 2021, signaling investor confidence in the company’s growing presence in the AI sector.
A New Flagship in the Qwen Line
Qwen3-Max is the newest flagship of Alibaba’s Tongyi Qianwen (Qwen for short) series, which translates to “seeking truth by asking a thousand questions.” The family first appeared in 2023 and has had several iterations, including the Qwen2 and Qwen2.5 releases that introduced mixture-of-experts architectures and instruction tuning. The Qwen3 line, launched earlier this year, added hybrid reasoning modes that allow users to toggle between “thinking” and “non-thinking” settings depending on task requirements.
Alibaba says Qwen3-Max is its largest and most capable model so far and is reported to contain more than one trillion parameters. Benchmarks shared by the company show Qwen3-Max achieving strong results on SWE-Bench, a measure of code problem solving, and Tau2-Bench, which tests how well a model can use external tools to complete tasks. On Tau2-Bench, the model outscored Anthropic’s Claude and DeepSeek’s latest releases, according to Alibaba.
Features That Stand Out
This graph shows results in popular coding and agentic benchmarks. (Source: Qwen.ai)
Several features distinguish Qwen3-Max from its rivals like Deepseek-V3.1 and GPT-5. The model retains the hybrid reasoning modes introduced with earlier Qwen3 systems, giving developers the option to run the model in a cost-efficient lightweight mode or a more resource-intensive reasoning mode. This flexibility is meant to appeal to enterprise users who need to balance accuracy with expense. Alibaba has also positioned Qwen3-Max as an agentic system, capable of carrying out multi-step tasks with less human prompting. The company says the model can handle tool use more reliably than earlier releases, an important capability for the agent-focused next stage of enterprise AI adoption.
Another key element is the model’s accessibility. Most of the Qwen line has been released under open licenses, and Alibaba has signaled that Qwen3-Max will continue this practice. By making the model available through APIs and cloud services, the company is trying to grow a developer ecosystem that could strengthen Alibaba Cloud’s position against both domestic and international competitors.
Alibaba's Investment and Expansion
The model launch coincided with Alibaba CEO Eddie Wu’s announcement that the company will exceed its earlier commitment of 380 billion yuan (about $53 billion) in spending on AI and cloud infrastructure over the next three years. Speaking at the company's annual conference in China, Wu said that demand for AI computing resources is arriving faster than expected and that Alibaba intends to meet that demand with larger data centers, expanded global capacity, and deeper integration of new tools.
Alibaba's U.S.-listed stock rose 8% to close at 176.44, a four-year high for the company. While stock moves often fade as initial enthusiasm subsides, the response could show how closely investors are watching for signals of credible AI capability from the major firms. For Alibaba, it shows that the company’s AI strategy can influence its market standing beyond the retail sector, mirroring another big player, Amazon.
For the global AI field, Qwen3-Max’s arrival raises questions about competition, access, and the geopolitics of AI. Chinese firms now have several models competing at the frontier, and they will shape how researchers, enterprises, and governments weigh issues of technology sovereignty and global competition. Already, Huawei is unveiling new compute clusters and pushing its next-generation AI chips, Tencent is eyeing expansion of Chinese open models overseas, and Baidu is quietly migrating model training onto its own hardware. The stakes are no longer just who builds the biggest model, but who commands the entire AI stack across borders.
