At the World Artificial Intelligence Conference in Shanghai, Alibaba introduced Qwen3.8, a 2.4 trillion parameter multimodal AI model it claims is the second largest globally, trailing only Anthropic’s Fable 5. However, the company provided no benchmarks or independent verification to substantiate this bold assertion.
- Qwen3.8 has 2.4 trillion parameters and handles images, video, documents, and text.
- Alibaba claims Qwen3.8 is second only to Anthropic's Fable 5 but offers no public benchmarks.
- The model is currently accessible via paid subscriptions with full open weights promised later.
What happened
Alibaba unveiled Qwen3.8 at the World Artificial Intelligence Conference in Shanghai, presenting it as the company’s most advanced AI model to date. The model boasts 2.4 trillion parameters and is their first trillion-parameter multimodal system, capable of processing images, video, documents, and text. Alibaba highlighted its strengths in coding, full-stack development, and data analysis.
The launch follows closely after Moonshot released the Kimi K3 model, intensifying competition among Chinese AI labs. Alibaba’s Qwen3.8 is available via the company’s Token Plan subscription and developer tools, with prices offered at a discount during the trial period. The promise of fully open weights has been made but without a clear timeline or licensing conditions.
Why it matters
Alibaba's claim that Qwen3.8 is the world’s second-largest AI model is significant as it signals China’s growing position in the global AI development race, especially as labs outside the U.S. ramp up investment and scale. Being multimodal and crossing the trillion-parameter threshold marks a technical milestone that could open new commercial and research opportunities.
However, the lack of published benchmarks, model cards, or third-party verification raises skepticism in the AI community. Unlike previous releases where Alibaba provided full evaluations, the absence of transparent data fuels uncertainty and limits immediate external validation of performance claims. This affects investor confidence and the broader industry’s ability to gauge the model’s true capabilities.
What to watch next
The broader AI field will closely watch when Alibaba releases Qwen3.8’s full open weights and licensing terms. Access to these will allow independent researchers and competitors to assess the model’s performance and verify its place on global leaderboards, such as those tracking coding or general multimodal tasks.
Alibaba’s strategic approach of initial paid access coupled with delayed open-weight release may influence how other developers balance commercialization and openness. As Chinese AI labs increasingly push scale and openness, this could reshape competition dynamics against established U.S. models and define the next phase of the AI arms race.