Z.ai ships GLM-5.3-Flash: a MIT-licensed open-weight flash model (320B-A18B) with 1M context and multimodal input that approaches Claude Opus 4.8 on coding and agentic benchmarks at a far lower price. Cloudflare · explainx.ai
Alibaba open-sources Qwen3.8-Flash-Next: an experimental open-weight MoE model (125B-A6B, only 6B active per token) that previews the next-generation Qwen4 architecture, released Aug 26. TechNode · Hugging Face
Korea’s Wrtn raises $72M Series C: the Seoul-based AI platform now values above $722M (over 1 trillion won), with its US-focused AI entertainment app OOC passing $7.2M in monthly revenue within three months of launch. The Korea Times
Self-driving truck startup Gatik raises $200M: led by Qatar Investment Authority and Koch Disruptive Technologies, its largest round to date, bringing total funding to roughly $500M. TechCrunch
Fitbit co-founders launch Luffu Link, a $250 AI health band: a screenless LTE wearable where background AI monitors activity, sleep and HRV, flagging changes and alerting family members. TechCrunch
India’s AI compute buildout accelerates: AM Intelligence ordered 9,000 Nvidia Vera Rubin NVL72 rack-scale systems — roughly $8B in capex — for a Hyderabad AI factory running on renewables. Taipei Times
Amazon shuts down Mechanical Turk on Sept 30: the 2005-era human microwork marketplace closes after 21 years, its demand hollowed out by AI-native labeling players like Scale AI and Mercor. CNBC
Prototype agents on cheap open-weight flash models: GLM-5.3-Flash’s 1M context and multimodal input make it a cost-efficient test bed for coding and agentic workloads. Cloudflare · explainx.ai
Gap for domain-specific human-in-the-loop data: the generic MTurk labor pool is gone, but specialized reliability-critical validation workflows (medical, legal) still need trustworthy human review. CNBC
China’s AI-chip IPO push: Shanghai’s Enflame opens subscriptions Sept 2 for a ~$892M STAR Market IPO, with heavy Tencent customer concentration flagged in the prospectus. Yahoo Finance