Alibaba’s Qwen research team has introduced Qwen3.8-Max, a new flagship multimodal AI model built around a 2.4-trillion-parameter mixture-of-experts architecture. The release positions the system as a major step forward for complex, autonomous computer-based tasks.
According to Alibaba, Qwen3.8-Max can independently handle software projects that run for more than 10 days. The company also says the model can reproduce research papers that require thousands of lines of code, highlighting its focus on extended technical workflows rather than short exchanges.
Qwen’s headline claim is that Qwen3.8-Max performs better in agentic computer use than GPT-5.6 Sol Max and Fable 5. Those comparisons are presented by the company as evidence of the model’s capabilities, while the available information does not provide independent verification of the results.
The launch adds another high-profile model to the competition over AI systems capable of planning, coding and completing multi-step tasks with limited human intervention. Qwen3.8-Max’s combination of multimodal processing and a large mixture-of-experts design is central to Alibaba’s positioning of the model.
