Chinese AI startup DeepSeek has made waves again with the quiet release of an upgraded version of its highly acclaimed R1 artificial intelligence reasoning model. The company, which rose to prominence earlier this year, did not make an official announcement for this update, dubbed R1-0528. Instead, the revised model became available on the AI model repository Hugging Face, its namesake chatbot website, and mobile apps, as communicated through a company-run WeChat group on Wednesday, May 29, 2025.

The original DeepSeek R1 model, launched in January 2025, had a significant impact on the global tech industry. This free, open-source reasoning model outperformed offerings from established rivals, including Meta and OpenAI, on several standardized metrics. What was particularly shocking was its low development cost, purportedly just several million dollars, and the short time frame of its creation. This success triggered a rout in global tech stocks, notably affecting AI stalwart Nvidia, as investors questioned the necessity for massive spending by U.S. tech giants on AI infrastructure.

The upgraded R1 model continues to be a reasoning model, designed to execute more complicated tasks through a step-by-step logical thought process. While DeepSeek did not disclose specific details of the changes in the R1-0528 update, it quickly garnered attention from the developer community. Independent benchmark platform LiveCodeBench reported that the new model demonstrates improved performance in AI-assisted coding. On the LiveCodeBench leaderboard, it ranks as the top Chinese model for coding capabilities, trailing only OpenAI’s o4-mini-high, o3-high, and o4-mini-medium, and outperforming models like Alibaba Group Holding’s latest Qwen3 and Anthropic’s Claude 3.7.

DeepSeek's advancements serve as a prominent example of China's continued development in artificial intelligence despite U.S. attempts to restrict the country's access to critical chips and technology. Other Chinese technology giants, such as Baidu and Tencent, have also revealed strategies to make their AI models more efficient in response to U.S. semiconductor export curbs. Underscoring this resilience, Nvidia CEO Jensen Huang recently criticized U.S. export controls, stating, "The U.S. has based its policy on the assumption that China cannot make AI chips. That assumption was always questionable, and now it's clearly wrong... The question is not whether China will have AI. It already does."

The debut and subsequent upgrade of the R1 model have turned DeepSeek's founder, Liang Wenfeng, into a tech celebrity and a symbol of China's capability to compete with Silicon Valley. His prominence was further highlighted in February 2025 when President Xi Jinping invited Liang to a high-profile gathering with some of China's most influential entrepreneurs, including Alibaba’s Jack Ma and Tencent’s Pony Ma. The announcement of the R1 upgrade also came shortly before Nvidia's latest financial report, a company whose shares had previously been impacted by R1's initial release. DeepSeek had last updated its foundational large language model, V3, in March 2025, touting improvements in coding and writing in that V3-0324 release on Hugging Face.