| Mobile Web

Search results for H200

AI & Enterprise

More than half of AI Safety Institute assessed models are open-weight, 58 percent Chinese

More than half of the AI models assessed by the AI Safety Institute this year were open-weight, it found. Of 34 unique models assessed from January to August, 19 were open-weight and 15 were closed. Chinese models accounted for 11 of the 19 open-weight models, including Kimi, DeepSeek and Qwen. The institute assesses risks such as prompt injection, data leakage and cyber exploits, but cited limited staff and computing resources.

AI & Enterprise

ETRI, KAIST take top spot in robot navigation challenge

South Korean researchers won first place in an international competition with robot AI that understands human speech and navigates to a destination on its own. The Electronics and Telecommunications Research Institute and the Korea Advanced Institute of Science and Technology formed a joint team that topped the VLNVerse challenge at the ECCV 2026 computer vision conference in Malmo, Sweden, beating 112 teams. The team recorded a 90.7 percent average success rate and won on earlier submission time.

AI & Enterprise

Naver Cloud to develop 700B cybersecurity AI model, preloads 4,000 GPUs

Naver Cloud said on Sept. 3 its consortium has been selected as the final operator for the government’s specialised AI foundation model project in the cybersecurity field and has begun full-scale development. It preloaded 4,000 B200 GPUs for pretraining ahead of the official start. The consortium will also use 256 government-provided GPUs and an additional 256 H200 GPUs from LG. It plans to develop two 700B MoE security models using HyperCLOVA X and LG AI Research’s EXAONE.