Alibaba Previews Qwen3.8 Max AI Model in Shanghai

Alibaba Group Holding Ltd. previewed its flagship multimodal AI model, Qwen3.8 Max, during the World AI Conference in Shanghai on July 20, 2026. The model features 2.4 trillion parameters and a 1 million token context window, making it the first in the Qwen family above 1 trillion parameters capable of processing text, images, video, and documents. Alibaba claims the system is comparable to leading frontier models and second only to Anthropic PBC's Fable 5, though the company has not yet released public benchmarks, a model card, or a license to support this ranking.

Technical evaluations show that Qwen3.8 Max excels in front-end development, specifically regarding SVG animations and responsive web layouts, as well as early-stage 3D game prototyping. However, the model exhibits slower output speeds than competitors and struggles with high-detail visual polish. This release follows the recent unveiling of Kimi K3, a 2.8 trillion-parameter model from Moonshot AI, as Chinese developers increasingly compete for parity with U.S. AI leaders.

The preview is currently accessible at a discounted rate via Alibaba's Token Plan subscription service and coding platforms, including Qoder and QoderWork. Alibaba intends to make the model open-weight in the near future and has announced plans for a Qwen 4.0 series to improve output quality and processing speed.