Nature, Published online: 23 January 2025; doi:10.1038/d41586-025-00229-6

DeepSeek-R1 performs reasoning tasks at the same level as OpenAI’s o1 — and is open for researchers to examine.