The video reviews Alibaba's claims for Qwen 3.8 Max, a large mixture-of-experts modelA mixture-of-experts model contains multiple specialized subnetworks and activates a selected subset for each input instead of using every parameter every time. positioned near leading closed systems on research, professional-work, visual-reasoning and computer-use evaluationsComputer use lets an AI agent interact with software through interface controls such as screens, buttons, keyboards and pointers.. Independent arena rankings are presented as partial support for that position, although the video notes that benchmarksA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions. do not capture the whole user experience.
Alibaba's longer demonstrations are the more important evidence in the argument. In separate controlled environments, the model reportedly maintained a software project for more than two weeks, reproduced and extended a research paper over several days, and repeatedly optimized an existing chip design. The environments and tools were engineered by peopleAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction., but the model handled extended cycles of proposing, testing and revising workAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior..
The strategic conclusion is that capable open models may turn the model layer into a commodity. If useful intelligence becomes inexpensive and portable, value can shift toward cloud capacity, chips, agent platforms and the applications built around the models. China can therefore compete through an integrated ecosystem even when its individual models are not always the absolute leaders.
Watch on YouTube




