DeepSeek is a family of models from the Chinese lab DeepSeek (Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd.), spanning a general-purpose V-series and a reasoning-focused R-series. The DeepSeek site lists models including DeepSeek V3 and DeepSeek R1, plus earlier and specialized lines such as DeepSeek Coder V2, DeepSeek VL, DeepSeek Math, and DeepSeek LLM, with links to public GitHub repositories.
The DeepSeek-R1 paper, “DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning” (submitted January 22, 2025; later published in Nature), shows that reasoning ability can be developed through reinforcement learning without large amounts of human-annotated reasoning data, producing behaviors such as self-reflection and verification.
As of June 2026, the current flagship is DeepSeek-V4. DeepSeek’s API docs announced the DeepSeek-V4 preview on April 24, 2026, shipping two models - DeepSeek-V4-Pro (1.6T total / 49B active parameters) and DeepSeek-V4-Flash (284B total / 13B active parameters) - both with a 1M-token context window and released as open weights on Hugging Face. The earlier V3 and R1 models remain listed. Specific current versions evolve, and the site links each model to its repository.
Why business readers should care: DeepSeek demonstrated that strong reasoning models can be built and released openly outside the largest US labs, a development that reshaped cost expectations and the competitive map for the whole industry.