Alibaba unveils Qwen3.8-Max with 2.4T parameters as US–China AI race heats up

- Qwen3.8-Max claims benchmark wins and open weights amid China–US AI rivalry - Alibaba stock surges over 7% as developers and investors eye preview and open access On July 21, 2026, Neowin reported that Alibaba unveiled Qwen3.8-Max, its most advanced AI model with 2.4 trillion parameters. According to Neowin, the company promised to provide open-weight access for developers within a week. This release comes as China–US competition in AI escalates, and Alibaba is leveraging strategic openness and performance claims to gain an edge in global markets. Qwen3.8-Max operates as a multimodal system that processes text, imagery, and video. The model uses a Mixture-of-Experts architecture and activates about 95 billion parameters per request to improve efficiency. According to Neowin, Alibaba’s internal benchmarks indicated that Qwen3.8-Max outperformed top competitors, including Anthropic’s Fable 5 and OpenAI’s GPT-5.6 Sol, in coding, agentic, and vision tasks, especially long-horizon reasoning. However, full public benchmarks and independent technical validation are still pending. Developers can therefore access a preview of Qwen3.8-Max via Alibaba Cloud’s API services and QwenWork. In addition, Neowin noted that, unlike US rivals’ proprietary approaches, Alibaba’s open-weight release has driven significant enterprise and developer interest. Alibaba positions Qwen3.8-Max just below Anthropic’s Fable 5 in performance rankings, while major evaluation platforms have yet to independently list or verify the model. Alibaba’s open-weight strategy and benchmark claims have intensified the China–US AI technology race, and the announcement drove its stock up more than 7% while fueling interest from both investors and enterprises. As a result, the preview launch and upcoming open distribution represent a decisive shift that may accelerate AI adoption in China and abroad. Industry observers note that Qwen3.8-Max enters the market as a claimed top performer with developer preview access, although they also point out that final technical documentation and third-party performance validation are still awaited.