Alibaba Unveils Qwen3 AI Models, Marking a New Milestone in China’s Open-Source Tech Push

Alibaba Group has released the latest version of its open-source large language models (LLMs), dubbed Qwen3, in what experts describe as another major step forward for China’s artificial intelligence (AI) sector.

The launch underscores the growing sophistication of Chinese AI labs and the increasingly competitive nature of the global open-source AI ecosystem.

The new Qwen3 series introduces several key upgrades over previous iterations, including enhancements in reasoning, instruction following, tool usage, and multilingual capabilities. According to Alibaba, these improvements position Qwen3 to rival other top-performing models, such as DeepSeek’s R1, across several industry benchmarks.

A notable innovation in Qwen3 is the inclusion of Alibaba’s first “hybrid reasoning models.” These models are designed to alternate between “thinking” and “non-thinking” modes—optimizing performance for tasks that range from complex programming to more routine, general-purpose responses. This flexible approach, the company says, allows for more efficient AI-powered application development across devices, including mobile platforms.

Qwen3 also features multilingual support for over 100 languages and dialects, and is available in eight model variations to suit a range of technical and commercial use cases. The models are openly accessible via platforms such as Hugging Face, GitHub, and Alibaba Cloud, and are already integrated into Alibaba’s AI assistant, Quark.

The company claims that one model in the new series—the Qwen3-235B-A22B MoE—significantly reduces deployment costs compared to other state-of-the-art alternatives, reflecting a broader goal to make powerful AI tools more accessible to developers and businesses.

AI experts have highlighted Qwen3’s release as a sign of China’s accelerating momentum in the AI field.

“This is a significant breakthrough—not just for performance but for the application potential,” said Wei Sun, principal AI analyst at Counterpoint Research.

Sun pointed to Qwen3’s hybrid reasoning, open-source nature, and multilingual capabilities as key differentiators.

Alibaba’s move comes amid intensified competition within China’s tech landscape. Domestic firms such as Baidu have also introduced new LLMs in response to the success of DeepSeek’s open-source R1 model earlier this year, which helped trigger a broader industry shift toward openness and cost-efficiency in AI development.

Despite ongoing pressure from US export controls, China’s AI labs continue to demonstrate their capacity to produce competitive technologies. Ray Wang, an analyst based in Washington, noted that the gap between US and Chinese AI capabilities is narrowing.

“Some would say the difference is now measured in months—or even weeks,” Wang commented.

Alibaba reports that its Qwen model series has already seen wide adoption, with over 300 million downloads globally and more than 100,000 derivative models created by the open-source community. With Qwen3, the company aims to further cement its role in both domestic and international AI development.

CNBC and Reuters contributed to this report.