Alibaba has unveiled its largest AI model to date, previewing Qwen3.8 Max during the World AI Conference in Shanghai. The company states that the system comprises 2.4 trillion parameters and described its performance as “second only to Fable 5,” Anthropic’s flagship model. The announcement arrived just two days after rival Moonshot AI introduced its own large-scale model, Kimi K3, which carries 2.8 trillion parameters.
Limited Transparency and Unverified Claims
Unlike previous Qwen releases, the preview comes without independent verification materials. No benchmark table, model card, or license has been published, and Alibaba has not disclosed how many parameters are active per inference request—a key figure that determines the real operating cost of a mixture-of-experts architecture. All performance assertions therefore originate solely from Alibaba. Qwen developer Shuai Bai shared additional context on social media, noting that Qwen3.8 is the family’s first multimodal model above one trillion parameters, capable of processing text, images, video, and documents.
Restricted Access and a Pricing Signal
Early testing is available, but only through Alibaba’s subscription-based Token Plan and the developer platforms Qoder and QoderWork. During the trial phase, the company is charging one-tenth of its standard price. Alibaba has also signaled plans to release open weights, though it has provided neither a timeline nor a license. If the company follows through, such a move would mark a departure from its usual practice of keeping the largest Max models closed to the public.
Practical Realities and Competitive Pressures
Even with open weights, immediate everyday impact would be limited. A 2.4-trillion-parameter model demands well over a terabyte of memory even under heavy compression, placing it beyond the reach of virtually all private hardware. The broader significance of the announcement lies in what it reveals about market demand and ecosystem strategy. Over the weekend, Moonshot AI was forced to halt new Kimi K3 subscriptions after its GPU infrastructure proved insufficient to handle the surge of interest. In a more tangible development, Alibaba’s compact derivative model, Bonsai 27B—built on Qwen3.6 technology—can now run directly on an iPhone, illustrating how the wider model family feeds into usable, device-ready tools.
Sources: x.com, www.scmp.com, www.marktechpost.com