Alibaba Releases Multimodal AI Model 'Qwen 3.8'

A new model 'Qwen 3.8' has been added to Alibaba's AI model series 'Qwen.' This is a multimodal model with 2.4 trillion parameters—a metric indicating the scale of an AI model's training—and is designed to handle multiple types of information beyond just text. It is currently available as a preview version.

The Qwen team positions the model's performance as 'second place after Fable 5.' The team self-evaluates it as being at a level comparable to competing major models, with open-weight format—where model weights are disclosed—serving as another defining characteristic. However, it is important to note that this is Alibaba's own evaluation, not a verification result from an independent third-party organization.

As context, China's AI development competition has intensified particularly in recent years. The Qwen series has consistently released multiple models in open-weight format, building a track record of adoption within the global developer community. Qwen 3.8 is reportedly in competition with China-based AI model 'Kimi K3,' offering a glimpse into the competitive landscape surrounding technological superiority both domestically and internationally.

The open-weight approach stands in contrast to closed models premised on commercial use. Because researchers and developers can more easily run and customize models in their own environments, it creates opportunities to promote technology dissemination and external improvements. Alibaba's continued pursuit of this strategy can be seen as a strategic intent to increase its influence within the developer community.

The parameter scale of 2.4 trillion ranks among the larger models currently available. However, a model's actual capabilities are not determined by parameter count alone, but involve numerous factors including training data quality, inference efficiency, and the degree of optimization for specific tasks. Going forward, its true capabilities are expected to become clearer through independent benchmark evaluations and feedback from actual users.

Alibaba's release of Qwen 3.8 represents a development indicating that large-scale AI model development extends beyond a handful of Western companies. Key points of future attention will be whether the open-weight approach gains acceptance from developers worldwide and how the performance gap relative to top-tier models like Fable 5 is ultimately evaluated.