2025-01-22
arXiv

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

Zhiyu Wu , Xiaokang Chen , Zizheng Pan , Xingchao Liu , Wen Liu
The paper introduces DeepSeek-R1-Zero, a model trained with reinforcement learning that exhibits strong reasoning capabilities but faces readability and language mixing issues. To improve these aspects, DeepSeek-R1 is developed, which uses multi-stage training and cold-start data, achieving performance on par with OpenAI-o1-1217. The models and additional resources are open-sourced.
We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoning capabilities. Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors. However, it encounters challenges such as poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates multi-stage training and cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1-1217 on reasoning tasks. To support the research community, we open-source DeepSeek-R1-Zero, DeepSeek-R1, and six dense models (1.5B, 7B, 8B, 14B, 32B, 70B) distilled from DeepSeek-R1 based on Qwen and Llama.