DeepSeek is a Chinese AI lab that has produced a series of models (DeepSeek V3, R1) that match or exceed GPT-4-level performance at a fraction of the API cost. The R1 model is particularly notable for transparent chain-of-thought reasoning. DeepSeek's models are open-weight and can be self-hosted, making them attractive for privacy-conscious developers. The web chat is free, and the API is among the cheapest available for frontier-class models.
Stan cen: March 2026