Alibaba Releases Qwen3.8-Max Model with 2.4 Trillion Parameters

Alibaba unveiled Qwen3.8-Max, the most powerful model in its Qwen series with 2.4 trillion parameters. Built on the Qwen3.5 architecture, the system performs complex long-duration tasks with minimal human intervention, demonstrating advanced capabilities in programming and multimodal agents. In practical tests, the model completed real-world two-week projects autonomously. A notable achievement was creating an autonomous agent framework from a single prompt, a task requiring 16 days of computation. Applications include compressing weeks of legal analysis into hours and developing end-to-end trading strategies. For extended-duration tasks, Qwen3.8-Max exhibits autonomous planning and closed-loop adaptive learning. The system processes 200-page documents, organizes 100-hour videos into searchable structures, and automatically detects errors during execution. On Arena benchmarks, Qwen ranks second only to Anthropic's Claude series among top-tier global models. The API is available through the Qwen AI platform at competitive pricing. Model weights will be open-sourced next week.