Chinese AI startup DeepSeek quietly released an upgraded version of its R1 reasoning model on Hugging Face, according to CNBC. The update builds on its earlier success that shocked markets by outperforming models from OpenAI and Meta. The improved model ranks just behind OpenAI’s o4-mini and o3 on LiveCodeBench. Rlesearchers say it demonstrates stronger math, coding, and reasoning capabilities while reducing hallucinations. DeepSeek exemplifies how Chinese AI development continues despite U.S. efforts to limit access to advanced technology. Baidu and Tencent are also improving model efficiency to adapt to U.S. semiconductor export restrictions.
