The 2.4 trillion-parameter model debuted at the World Artificial Intelligence Conference in Shanghai, with open weights promised soon
NurPhoto / Getty Images
Alibaba Group previewed its new flagship AI model, Qwen3.8, at the World Artificial Intelligence Conference in Shanghai on Sunday, claiming the system is second only to Anthropic's Claude Fable 5 among the world's frontier models. Alibaba stock rose as much as 5.4% on Monday following the announcement.
The preview build, called Qwen3.8-Max-Preview, carries 2.4 trillion parameters — a milestone that puts it among the first models in the Qwen family to cross the trillion-parameter threshold while also supporting images, video, and documents, the company said. Alibaba described the model as "continuously evolving" and said it plans to make it available as open weights, meaning developers could download and modify it, though no date or license terms have been set.
During the trial period, the preview is priced at a tenth of normal rates and is accessible via Alibaba's Token Plan subscription as well as its Qoder and QoderWork developer platforms, the company said.
The performance claim, however, comes without supporting evidence. The preview arrived without any supporting technical documentation — no benchmark data, no model card, and no activated-parameter count, according to SiliconANGLE. No independent leaderboard has scored Qwen3.8. On LMArena, Fable 5 holds the top position, while its predecessor Qwen3.7-Max placed considerably lower on the same ranking. By contrast, Qwen3.7-Max launched in May accompanied by comprehensive documentation, among them a score of 56.6 on the Artificial Analysis Intelligence Index.
The launch comes three days after Beijing-based Moonshot AI released Kimi K3, a 2.8-trillion-parameter open model that drew attention in Silicon Valley and prompted Moonshot to pause new subscriptions to manage demand. Together, the two releases signal that Chinese developers are now operating at parameter counts that had been the exclusive domain of top American AI laboratories, according to SiliconANGLE.
Alibaba has made its Qwen model family central to its AI ambitions. The company received regulatory clearance in China to power Apple $AAPL Intelligence across iOS, iPadOS, macOS, and visionOS for users in the country. Alibaba has also committed to spending more than 380 billion yuan ($53 billion) on cloud and AI infrastructure over three years, and earlier this year unveiled its Zhenwu M890 AI chip, built by its T-Head semiconductor subsidiary, as part of an effort to build domestic computing infrastructure amid U.S. export restrictions on advanced chips.
The open weights, once released, would allow outside researchers to run independent evaluations of Qwen3.8's capabilities. Until then, the claim that it ranks second globally remains Alibaba's own.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
