Number of the week: $200 million—the funding round closed by world-model startup VAST the same week China registered a national World Model Evaluation Standard for drafting.
Full issue, with sources, tables, and paper highlights:
TC28/SC42 also began drafting 25 new AI standards: five on embodied intelligence (including an ethics governance guide), three on AI for Science, a world-model evaluation spec, and an AI Bill of Materials format. CAC published four expert interpretations of TC260-005.
Notable models: MiniMax released M3, its first 2026 flagship—a 1M-context agentic foundation claiming SWE-Bench Pro 59.0—paired with MiniMax Code, a downloadable desktop coding agent. Also releases from Alibaba, Tencent, Moonshot, and StepFun.
What happened in Chinese AI governance, development, and safety in the last few weeks? Read on for highlights of Issue 5 of the China AI Bulletin 🧵👇
On governance: Qiushi published Xi Jinping's January Politburo speech on future industries. Xi calls for prospectively addressing risks like "loss of control of technology, ethics anomalies, and data misuse"—the first time 技术失控 appears in Xi's published speeches.
China's standards body approved GB/Z 185, a seven-part agent-interconnection family covering identity, discovery, description, interaction, and tool invocation.
On international governance: MIIT and ASEAN inaugurated the China-ASEAN AI Industry Innovation Center in Beijing, with a standards-and-governance mandate built in from inception. State-visit readouts with Serbia and Pakistan both name AI as a designated cooperation area.
SAMR and NDRC issued China's first AI metrology guidance, pledging measurement infrastructure to make AI performance "measurable, comparable, and traceable." It names algorithmic black boxes as a pain point and calls for R&D on monitoring AI systems' internal states.