DeepSeek Updates Inference AI Model 'R1', Improves Coding Performance

Details of the Update Not Disclosed

Chinese artificial intelligence (AI) startup DeepSeek has updated its inference AI model, 'R1'.

According to the South China Morning Post (SCMP) and other sources on May 29, DeepSeek announced in a notice posted on social media platform WeChat the previous day that "a minor update to the R1 model has been completed." However, while the update was named 'R1-0528' using the announcement date, the company did not disclose specific details of the update.

The updated model is currently available on the open-source AI platform Hugging Face.

SCMP reported that although the specific details of 'R1-0528' have not been made public, the update quickly drew attention in the developer community. Independent benchmarking platform LiveCodeBench stated that the updated R1 model has shown improved performance in the coding field. In LiveCodeBench's coding performance evaluation, 'R1-0528' ranked second, following OpenAI?the developer of ChatGPT?with its models 'o4-mini-high', 'o3-high', and 'o4-mini-medium'.

Previously, DeepSeek attracted global attention when it launched 'R1' on January 20. At the time, DeepSeek stated that although development costs were much lower than those of major U.S. tech companies, benchmark evaluations showed that its inference model outperformed OpenAI's 'o1' in mathematics and coding.

Industry observers expect DeepSeek to unveil its next inference model, 'R2', in the near future.

This content was produced with the assistance of AI translation services.