DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

arXiv:2501.07563

DeepSeek-AI

We introduce DeepSeek-R1, a reasoning model trained with large-scale reinforcement learning that achieves strong multi-step reasoning without extensive human-annotated chain-of-thought data. The model demonstrates competitive performance on mathematics, coding, and scientific reasoning benchmarks through outcome-based RL and multi-stage training.