DeepSeek R1 is the open-weight reasoning model that shocked the world — matching o1 on math, code, and reasoning benchmarks while being released under MIT license. Free to use at chat.deepseek.com.