Short briefs on the AI industry, built from local-language reporting.
Why it matters
Utopai's benchmark result shows that domain-specific post-training for film production can elevate an open-source foundation model past dedicated AI labs without pre-training from scratch.
The catch
The studio's $1 billion valuation and its projected $110 million in pre-sale revenues are external estimates reported by Forbes rather than audited financial disclosures.
StartLux open-sources decision models outscoring Jev on benchmark index
AMD agrees to acquire World Labs for $8.2 billion
Honor ships Magic9 with Alibaba agent hitting 90% completion
AgiBot-backed Qiyuan prices home humanoids at about $3,000
StepFun releases 600-billion-parameter Step 5 Preview model
DeepSeek details system running 3 million daily agent sandboxesOur read on recent briefs, in one line each
Training data gathered by teleoperating real robots is limited by how many robots and operators there are.MiFeng pays people to record chores for robot training 
China is setting an official yardstick for "AI devices" built on agent behaviour — understanding intent, splitting tasks, operating apps — rather than on the chip or model inside, and plans to use it in consumer subsidies.Lenovo tablet reaches top rung of China’s AI-device grades 
The competitive advantage is shifting from those who can generate content to those who can govern execution.How Risk Topology Shapes AI System Design 
