Alibaba Unveils Qwen 3.8 With 2.4 Trillion Parameters, Claims Frontier AI Performance

Alibaba's Qwen team has released Qwen 3.8, a 2.4-trillion-parameter open-weight model that it says matches frontier AI systems and trails only Anthropic's Fable 5. The model is multimodal, processing images, video, and documents, and is available at 10 percent of standard pricing.

Alibaba has unveiled Qwen 3.8, its most powerful open-weight AI model to date, with 2.4 trillion parameters and multimodal capabilities spanning images, video, and document processing. The Qwen team says the model matches frontier performance and trails only Anthropic's Fable 5.

According to The Decoder, developer Shuai Bai confirmed this is the team's first multimodal model exceeding one trillion parameters. A preview is available through Alibaba's Token Plan, Qoder, and QoderWork at 10 percent of standard pricing. Open weights are promised soon, though no benchmark results have been released yet.

The launch is strategically timed. Qwen 3.8 arrives just as rival Moonshot AI rides momentum from its Kimi K3 model. Moonshot, which reached 00 million in annual recurring revenue in June, is reportedly planning an IPO within six months. Qwen's open-weight release could disrupt Moonshot's strategy of funneling interest through its chat app and API into paying customers.

The model builds on Qwen's increasingly credible track record. The Qwen project on GitHub has become one of the most active open-source AI repositories, part of a broader shift that has seen Chinese open-weight models close the gap with Western frontier systems. With 2.4 trillion parameters open-sourced, Qwen 3.8 continues the trend of making state-of-the-art AI capabilities broadly available.