AI Industry Roundup: ByteDance’s 5-Trillion Parameter Model Plan, Alibaba’s Office AI Launch, and Key Global Developments

In a significant AI industry move, ByteDance is reportedly exploring the development of a large language model with over 5 trillion parameters—outpacing Alibaba’s Qwen 3.8-Max (2.4 trillion) and Moonlight AI’s K3 (2.8 trillion), making it the largest known model in China. Led by Xiang Liang (Seed Foundation head) and Shen Ke (LLM pre-training data lead), the project is still in early stages and may not be released publicly.

Alibaba has merged three independent Agent lines—QoderWork (programming), MuleRun (task scheduling), and Wukong (desktop interaction)—into a new product named "Qianwen Office", managed by Chen Yusen, the 34-year-old who took over DingTalk in June.

Google’s market value plummeted by $1800 billion in one day following DeepMind leadership shifts: Demis Hassabis (2024 Nobel Chemistry Prize winner) stepped down as CEO to become DeepMind chairman and Alphabet chief scientist, while Koray Kavukcuoglu was promoted to senior VP reporting directly to Sundar Pichai.

Elon Musk stated that Tesla’s Optimus humanoid robot will consume about 25% of Terafab’s AI computing output, with the remaining 75% allocated to space AI projects.

Hong Kong research teams made notable advances: a joint study by Hong Kong Baptist University, University of Hong Kong, and BNU-HKBU UIC revealed blind spots in automatic fact-checking systems (published at ACM MM 26). Additionally, Hong Kong Baptist University developed the VisualPatchWorld framework to help AI understand physical world dynamics (arXiv preprint, July 2026).

Sources

• Sina Tech: ByteDance’s 5-trillion parameter model plan
• Wuxiang Business Trends: Alibaba’s Qianwen Office launch
• First Financial: Google’s market cap drop and DeepMind changes
• Sina Finance: Musk on Tesla Optimus computing allocation
• Tech Walker: AI fact-checking blind spots research
• Tech Walker: VisualPatchWorld framework for physical world AI