DeepSeek · model release · downloadable model
DeepSeek-R1-0528
DeepSeek updates its R1 reasoning model with better results and fewer made-up answers.
In detail
DeepSeek released R1-0528, an updated R1 with improved benchmark performance, better front-end capabilities, reduced hallucinations, and JSON output and function-calling support. API usage is unchanged and weights are on Hugging Face.
Sources
- DeepSeek-R1-0528 Release · DeepSeek · Primary source
Last checked 11 November 2026.