Chinese artificial intelligence startup DeepSeek has quietly rolled out a significant upgrade to its R1 reasoning model, intensifying the global AI race and challenging the dominance of U.S. firms like OpenAI.
The updated model, R1-0528, was published on Thursday morning via developer platform Hugging Face, though no official public announcement has been made by the company.
Despite the low-key release, the model has quickly drawn attention after its performance was recorded on the LiveCodeBench leaderboard — a respected code generation benchmark developed by researchers from UC Berkeley, MIT, and Cornell.
R1-0528 ranked just behind OpenAI’s o4-mini and o3 models but outperformed notable competitors such as xAI’s Grok 3 Mini and Alibaba’s Qwen 3, marking a significant leap for China’s AI capabilities.
According to a Bloomberg report, a DeepSeek representative shared in a WeChat group that the rollout was a “minor trial upgrade,” inviting developers to begin testing the model. The company has not yet published detailed technical specifications or official performance comparisons.
The quiet launch comes months after R1’s debut in January, which shook global tech markets and challenged the long-standing assumption that cutting-edge AI requires massive U.S.-dominated compute infrastructure. DeepSeek’s efficiency-first approach has made waves by delivering comparable performance to U.S. models at a fraction of the computational cost.
Since then, Chinese tech giants such as Alibaba and Tencent have responded with their own high-performing models, and international players like Google and OpenAI have slashed prices and launched smaller, more efficient AI variants, such as OpenAI’s o3 Mini.
DeepSeek is also expected to release its more advanced R2 model, a planned successor to R1. Reuters reported in March that the R2 release was initially scheduled for May, though no official update has followed.
In March, DeepSeek also upgraded its V3 large language model, reinforcing its ambition to become a leader in both reasoning and general-purpose AI.
