Alibaba’s Qwen team previewed Qwen3.8-Max, a sparse mixture-of-experts model with 2.4 trillion total parameters and the team’s first multimodal system above 1 trillion parameters, able to process text, images, video and documents (MarkTechPost, The Decoder). The preview is reachable through Alibaba’s Token Plan, Qoder and QoderWork at 10 percent of standard pricing, and the company says the model should outperform Qwen3.7-Max on coding and on longer productivity work such as full-stack development and data analysis (The Decoder).
Alibaba describes the model as second only to Fable 5 among the systems it benchmarked, but no benchmark results have been released, and the model card and license are also outstanding (MarkTechPost, The Decoder). Open weights are promised soon with no date or license attached, and the weights are not yet posted to Hugging Face. Active parameters per token also remain undisclosed, which leaves the deployment cost unclear: at 4-bit precision a 2.4 trillion-parameter model needs roughly 1.2 terabytes for weights alone (MarkTechPost).