Full Breakdown
Alibaba Unveils Qwen3.8-Max, Its Largest AI Model Yet
8/3/2026, 11:58:29 AM
Alibaba Launches Qwen3.8-Max, Its Largest AI Model
On August 3, Alibaba Group Holding announced the rollout of Qwen3.8-Max, a multimodal foundation model built on 2.4 trillion parameters. The company described the architecture as a “mixture-of-experts” design that activates only specialized components for each request, aiming to lower computational costs and latency.
Capabilities, Scale and Benchmark Performance
Qwen3.8-Max can process text, images and video and handles up to 1 million tokens in a single context window, enabling analysis of lengthy documents, software codebases or multi-hour video streams. On the Arena AI leaderboard, it achieved the top spot among Chinese text models and placed second globally on the image-analysis ranking, trailing only a Claude Fable 5 variant. Business Times reported that the model’s benchmark scores are “comparable or sometimes better” than Anthropic’s Fable 5, a leading U.S. system that has faced export controls.
Access Plans and Upcoming Open-Weight Release
Alibaba made Qwen3.8-Max immediately accessible to developers worldwide through Alibaba Cloud’s Model Studio APIs and the QwenWork platform, a productivity suite positioned against Tencent’s WorkBuddy and overseas rivals such as Claude Cowork and ChatGPT Work. The firm also said it will publish the model’s weights for public download in the week beginning Aug 10, allowing users to customize the technology.
Market Reaction and Competitive Landscape
Following the announcement, Alibaba’s Hong Kong-listed shares jumped up to 7.3 %, the strongest rise in nearly a month. The launch comes amid a rapid escalation among Chinese AI developers: Moonshot AI’s Kimi K3, unveiled last month, carries 2.8 trillion parameters, while DeepSeek expanded access to its V4 Flash model on July 31. Analysts note that the narrowing gap with U.S. labs reflects intensified competition for talent, data and compute resources.
Verbatim Quotes
- “Many investors continue to underestimate Chinese AI models because of US chip restrictions or general scepticism,” — Ling Vey-Sern, managing director at Union Bancaire Privee
