DeepSeek R1-0528 Spurs US AI Competition

Chinese AI startup DeepSeek has launched its updated R1-0528 reasoning model on Hugging Face, showing enhanced mathematical, programming, and logical skills with fewer hallucinations compared to previous models. R1-0528 now ranks just behind OpenAI's o4 mini and o3 models on LiveCodeBench, outperforming xAI's Grok 3 mini and Alibaba's Qwen 3. The public release, despite lacking detailed documentation or a formal announcement, has drawn significant interest from the developer community. The strong performance and affordability of R1-0528 have prompted US tech companies to react by lowering prices and improving their own offerings. DeepSeek’s progress underscores China’s growing influence in advanced AI amid ongoing US export controls. The company is expected to release an even more advanced R2 model later this year.