Chinese AI startup DeepSeek has quietly rolled out an update to its R1 reasoning model, intensifying competition with American players like OpenAI. The revised model, named R1-0528, was uploaded to the developer platform Hugging Face early Thursday. However, the company has yet to issue a formal public statement or share detailed documentation or performance benchmarks.
Despite the low-profile release, the updated model has already made waves. According to the LiveCodeBench leaderboard, a respected benchmark developed by researchers from UC Berkeley, MIT, and Cornell. DeepSeek’s latest version ranks just below OpenAI’s o4 mini and o3 models in code generation performance, while outperforming xAI’s Grok 3 mini and Alibaba’s Qwen 3.
Bloomberg was the first to report the update on Wednesday, citing a message shared in a WeChat group by a DeepSeek representative who described the release as a “minor trial upgrade” and encouraged users to begin testing it.
Earlier this year, DeepSeek disrupted the prevailing belief that U.S. export restrictions would significantly hinder China’s AI progress. The company unveiled AI models that rivaled or even surpassed leading U.S. systems offered at a much lower cost. The original R1 release in January sent tech stocks tumbling outside of China and challenged the notion that large-scale AI development is only possible with massive computational resources and budgets.
Following R1’s debut, Chinese tech powerhouses including Alibaba and Tencent have launched models that claim to outshine DeepSeek’s offering. Meanwhile, global players like Google’s Gemini and OpenAI have responded by adjusting pricing strategies. Gemini with discounted access tiers and OpenAI with a more lightweight o3 Mini.
Industry watchers continue to anticipate the release of R2, DeepSeek’s next-generation model, which Reuters previously reported was initially scheduled for a May launch. The company also released an enhancement to its V3 large language model in March.
