DeepSeek quietly updates its R1 reasoning model

Chinese AI startup DeepSeek has released an update to its R1 reasoning model. The new version, named R1-0528, was published on developer platform Hugging Face on May 29, although the company has not yet issued an official announcement, as per a recent Reuters report.

While DeepSeek has not provided technical documentation or performance comparisons, the updated R1 model has appeared on the LiveCodeBench leaderboard, a benchmark designed by researchers from UC Berkeley, MIT, and Cornell, where it ranks just behind OpenAI’s o4 mini and o3 models in code generation tasks. It also scored ahead of Elon Musk’s xAI Grok 3 mini and Alibaba’s Qwen 3, adds the Reuters report.

The update was initially reported by Bloomberg on May 28, citing messages from a DeepSeek representative in a WeChat group. According to the message, the update is considered a “minor trial upgrade” and users have been invited to begin testing it.

DeepSeek gained attention earlier this year after launching the original R1 model, which challenged assumptions that US export controls would significantly limit China’s AI capabilities. The release of R1 in January sent shockwaves through the global tech industry, briefly affecting tech stocks outside China and prompting rival companies to respond with updated models and pricing changes.

Since then, major Chinese companies including Alibaba and Tencent have introduced AI models claiming performance improvements over DeepSeek’s. Meanwhile, US competitors such as OpenAI and Google have adjusted their strategies by releasing lighter models and offering more affordable access tiers.

DeepSeek is also expected to launch a more significant follow-up model, R2. According to a March report by Reuters, the company had originally planned to release R2 in May. In addition to the R1 update, DeepSeek released an upgrade to its V3 large language model earlier this year.