Chinese startup DeepSeek has released an update to its R1 reasoning model, increasing competition with OpenAI and other Western developments. Version R1-0528 is now available on the Hugging Face platform, Reuters reports.
The model improved its ranking in the LiveCodeBench ranking — a benchmark from researchers at UC Berkeley, MIT, and Cornell — where it was rated slightly below OpenAI o4 mini and o3, but above xAI's Grok 3 mini and Alibaba's Qwen 3.
A DeepSeek representative in a closed WeChat group called it a "minor test update" and said that users can already try it out.
The R1 model was first unveiled in January 2025. It attracted attention for its high efficiency at low cost, which challenged the US dominance in the industry. However, it turned out that DeepSeek was able to develop the model largely thanks to previous achievements by other companies. In response, other Chinese companies, including Alibaba and Tencent, began to demonstrate their own developments, which were intended to surpass DeepSeek.
DeepSeek is expected to soon introduce the next generation of AI — the R2 model.