DeepSeek, a Chinese artificial intelligence startup, increased competitiveness with US rivals like OpenAI by releasing an update to its R1 reasoning model early on Thursday.
Although DeepSeek has not yet made a formal public statement, they have released R1-0528 on the Hugging Face developer platform. Neither comparisons nor a description of the model were published.
However, DeepSeek’s upgraded R1 reasoning model was ranked just a little below OpenAI’s o4 mini and o3 reasoning models on code creation, and ahead of xAI’s Grok 3 mini and Alibaba’s Qwen 3 on the LiveCodeBench leaderboard, a benchmark created by academics from UC Berkeley, MIT, and Cornell.
The update had been reported on Wednesday. After completing what it called a “minor trial upgrade,” a DeepSeek representative reportedly informed a WeChat group that users could begin testing it.
Earlier this year, DeepSeek launched AI models that were on par with or better than industry-leading models in the US at a fraction of the price, shattering notions that US export restrictions were preventing China from making significant strides in AI.
When R1 was released in January, tech stocks outside of China fell precipitously, casting doubt on the notion that expanding AI calls for significant financial and computational resources. Since the introduction of R1, Chinese tech behemoths Tencent and Alibaba have produced models that they say outperform DeepSeek’s.
OpenAI lowered rates and unveiled an o3 small model that uses less processing power, while Google’s Gemini has introduced discounted access tiers.
A successor to R1, R2, is still highly anticipated to be released by the firm. R2’s release was originally scheduled for May, according to a March Reuters report that cited sources. Additionally, in March, DeepSeek announced an update to its V3 large language model.
