Alibaba on Monday unveiled Qwen 3.8-Max, the most powerful model in its Qwen family. The model, which has 2.4 trillion parameters but activates only 95 billion at a time, will be available via Alibaba Cloud's Model Studio, with the open weights to follow next week.
Qwen 3.8-Max sits close to the size of domestic rival Moonshot AI's Kimi K3, which launched last month with 2.8 trillion parameters. Both models can handle text, images, and video, and process up to one million tokens at a time. On the crowdsourced comparison platform Arena.AI, Qwen 3.8-Max immediately became the highest-ranking Chinese model for text tasks, though it still trails several Anthropic offerings including Claude Fable 5. For vision tasks, it ranked second globally, behind only a variant of Fable 5.
The model uses a mixture-of-experts architecture, which divides work among specialised segments rather than activating the entire system for every query. Alibaba said this keeps costs and response times lower. By contrast, OpenAI, Anthropic, and Google do not disclose parameter counts for their closed-source models.
Alibaba said Qwen 3.8-Max can complete complex, multi-day projects end-to-end. In one test, the model spent over ten days autonomously coding a self-evolving software harness from scratch, incorporating user feedback, running its own tests, and iterating through code, previews, and logs without human help. In another, it reproduced a machine learning research paper from zero, running 33 rounds of GPU training over roughly 125 hours, writing 7,600 lines of code, and then inventing 18 improvement ideas that ultimately outperformed the original paper's method. Entered into a real online contest with 526 human teams, it beat 87 per cent of the field.
The company also demonstrated the model's breadth across professional tasks, including a full legal compliance review completed in under an hour, an interactive banking app prototype delivered with no revisions, and a 26-dish restaurant menu with calorie and cost analysis produced in a single pass.
