Chinese artificial intelligence startup DeepSeek has quietly rolled out an upgraded version of its R1 reasoning model, escalating competition with U.S. leaders like OpenAI and Meta. The updated model, R1-0528, appeared without formal announcement on developer platform Hugging Face this week, where it quickly ranked among the top-performing models on LiveCodeBench, a benchmark designed by researchers from UC Berkeley, MIT, and Cornell.
LiveCodeBench scores placed the new DeepSeek model just behind OpenAI's o4-mini and o3 models in code generation and reasoning performance. It also outranked models from Elon Musk's xAI and Alibaba's Qwen 3, signaling China's continued push to close the AI gap despite restrictions imposed by Washington.
"The U.S. has based its policy on the assumption that China cannot make AI chips," Nvidia CEO Jensen Huang said Wednesday. "That assumption was always questionable, and now it's clearly wrong." He added, "The question is not whether China will have AI. It already does."
DeepSeek's R1 model initially launched in January, gaining attention for its open-source accessibility and surprisingly competitive performance relative to major U.S. offerings-despite being developed with far fewer resources. The release triggered a sharp sell-off in AI-related U.S. equities, including Nvidia, amid fears of price wars and lowered barriers to entry. Though U.S. markets later recovered, the episode forced a reassessment of the cost-efficiency assumptions underlying major AI deployments.
DeepSeek's R1-0528 upgrade was referred to internally as a "minor trial upgrade," according to a company representative who posted in a WeChat group. The update includes enhancements to reasoning capabilities, though no official documentation has been released.
The company's rapid development cycle has made it a focal point in China's broader AI race. Tech giants like Baidu and Tencent have also begun unveiling more efficient AI models tailored to work within the constraints of U.S. semiconductor export controls. In March, DeepSeek released an upgrade to its V3 large language model and was expected to unveil a successor, R2, as early as May, according to sources cited by Reuters.
OpenAI and Google have since adjusted their pricing strategies, offering smaller models such as the o3 Mini and discounted access to Gemini to compete with emerging low-cost challengers.
