Alibaba claims that Qwen3.8 is the second-best AI model in the world.

      Days after a competing Chinese laboratory stirred things up in Silicon Valley, Alibaba has responded. The company introduced Qwen3.8, its most advanced model to date, and boldly stated that it ranks only behind one model globally.

      Alibaba revealed this model at the World Artificial Intelligence Conference in Shanghai. The Qwen team announced on X that Qwen3.8 is “second only to Fable 5,” referring to Anthropic's latest version. As a result, Alibaba's stock rose by as much as 5.4% on Monday.

      What is Qwen3.8?

      The model features 2.4 trillion parameters, making it the first in the Qwen series to exceed a trillion and be multimodal, which means it can process images, videos, documents, and text simultaneously. Alibaba claims that Qwen3.8 will outperform its previous flagship, especially in coding and office tasks such as full-stack development and data analysis. Lead developer Shuai Bai referred to it as the team's first trillion-parameter multimodal model, asserting that its understanding either matches or surpasses that of leading proprietary systems.

      The latest updates from the EU tech scene, a narrative from our veteran founder Boris, and some dubious AI art are featured in a weekly newsletter. It’s free, so sign up now!

      A claim without evidence

      However, there’s a caveat. Alibaba did not provide benchmark scores, a model card, or independent testing to support the claim of being “second only to Fable 5,” as reported by SiliconANGLE.

      This marks a change for the company, as its previous flagship, Qwen3.7-Max, was launched in May with complete published results. In contrast, this new model was introduced without any data.

      This discrepancy is significant because the comparison is self-referential. No public leaderboard has assessed Qwen3.8. In LMArena, where Fable 5 holds the top position, the earlier Qwen3.7-Max is significantly lower on the list. The release of the weights would allow external researchers to validate the claim, but currently, they can't.

      The open-weight race

      The timing of this launch is strategic. It occurred just three days after Moonshot introduced Kimi K3, a 2.8-trillion-parameter open model that topped a major coding leaderboard and unsettled US chip stocks. According to The Decoder, Alibaba’s response is seen as direct.

      A group of Chinese labs is now releasing models with capabilities previously exclusive to the largest US labs. Moonshot has Kimi K3, while Zhipu has GLM-5.2. All are pushing the boundaries, with most offering their weights for free.

      There is a strategic nuance here. Like Moonshot, Alibaba is first releasing the model in paid products and planning to make the weights available later. This approach allows them to capitalize on the initial buzz and gain customers before anyone can access the model for free.

      The more significant signal is Alibaba's commitment to eventually releasing its weights. Historically, its largest models have not been open-weight, a trend noted by several analysts. AI researcher Nathan Lambert remarked that "something is changing," interpreting this shift as a new phase of competition among Chinese labs.

      If Alibaba's claimed number-two system can back up its assertion with matching weights, it would push the open-weight field further into advanced territory, an area the US labs have closely protected. However, the proof will only arrive once the download is available. Until then, Alibaba is asking the market to take its claims on trust.