Alibaba Group（阿里巴巴集团）has previewed Qwen3.8, its latest flagship artificial-intelligence model, saying the system has 2.4 trillion total parameters and will eventually be released with open weights.
The Qwen team announced the model on July 19. A preview version, Qwen3.8-Max-Preview, is already available through Alibaba's Token Plan and its Qoder and QoderWork coding platforms. The company has not disclosed a date or licence for the planned open-weight release.
Alibaba described Qwen3.8 as comparable with leading frontier models and said its internal assessment placed the system behind only Anthropic's Fable 5. The claim should be treated as the company's own evaluation: Alibaba had not published a full benchmark table or model card at the time of the announcement, and independent test results were not yet available.
The model is designed for multimodal tasks and can process text, images, video and documents. Alibaba said it also improves on the previous generation in coding, data analysis and complex office workflows. The company has not disclosed how many of the model's parameters are activated for each token, an important measure of the computing resources required to run a mixture-of-experts model.
The planned release would extend Alibaba's open-weight strategy around Qwen（千问）, one of China's most widely used families of large language models. Open weights allow developers to inspect, adapt and deploy a model on their own infrastructure, although they do not necessarily make the underlying training data or development process public.
Qwen3.8 arrives amid intensifying competition among Chinese AI developers. Its scale and promised open-weight release may broaden Alibaba's reach among overseas developers, but its performance and practical deployment costs will be easier to assess only after the company publishes the weights, technical documentation and reproducible evaluation results.