DeepSeek, which in late November unveiled DeepSeek-R1, an answer to OpenAI’s o1 “reasoning” model, is a curious organization.