Loading / 加载中
OpenAI has restricted the rollout of its GPT-5.6 model after a government request, saying such limitations should not be the norm. The news triggered speculation about cloud model reliance and boosted interest in local LLM alternatives. Some observers noted the development could benefit China's AI sector without direct effort.
SupraLabs has released SupraSafety-18M, a binary text classifier with 18 million parameters built on a BERT-style architecture. The model was trained from scratch using 2 T4 GPUs for 7 epochs on the nvidia/Nemotron-3.5-Content-Safety-Dataset. It achieves 81.2% accuracy and 86.9% precision in distinguishing SAFE vs. UNSAFE text. Designed for edge devices, mobile phones, and low-latency environments, it provides confidence scores, as demonstrated by example predictions. The model is available on Hugging Face.
A GPU modder operating a small lab in the US and working with Chinese factories warns that 96GB versions of the RTX 4090 and RTX 5090 are a scam as of June 2026. These cards do not exist, and no production is underway; sellers are exploiting the demand for high-VRAM local AI inference. The only recent high-capacity modded card the lab has received is a 32GB RTX 4080 Super, while 48GB RTX 4090 PCBs are in development. The modder urges the community to avoid any 96GB 4090 or 5090 offers.