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, demonstrated remarkable performance on reasoning.
With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.
However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,
we introduce DeepSeek-R1, which incorporates cold-start data before RL.
DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.
To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.
The R1 distilled models use specific system prompts to guide the model's behavior and responses. These prompts are designed to ensure that the model generates coherent and contextually appropriate outputs. Below are the system prompts used for the R1 distilled models:
1.**General Prompt**: "You are an AI assistant. How can I help you today?"
2.**Math Problem Prompt**: "Solve the following math problem step-by-step."
3.**Code Generation Prompt**: "Write a code snippet to accomplish the following task."
4.**Reasoning Task Prompt**: "Explain your reasoning for the following question."
### Expected Prompt Format
When interacting with the R1 distilled models, it is important to follow the expected prompt format to achieve the best results. Here are some examples of the expected prompt format:
1.**General Inquiry**:
- User: "What is the capital of France?"
- Model: "The capital of France is Paris."
2.**Math Problem**:
- User: "Solve the following math problem step-by-step: What is the derivative of x^2 + 3x + 2?"
- Model: "To find the derivative of the function x^2 + 3x + 2, we apply the power rule. The derivative is 2x + 3."
3.**Code Generation**:
- User: "Write a code snippet to reverse a string in Python."
- Model: "Here is a Python code snippet to reverse a string:\n```python\ndef reverse_string(s):\n return s[::-1]\n\nprint(reverse_string('hello')) # Output: 'olleh'```"
4.**Reasoning Task**:
- User: "Explain your reasoning for the following question: Why does the sun rise in the east?"
- Model: "The sun appears to rise in the east because the Earth rotates on its axis from west to east. As the Earth rotates, different parts of the planet are exposed to the sun's light, creating the appearance of the sun rising in the east and setting in the west."
**Post-Training: Large-Scale Reinforcement Learning on the Base Model**
- We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.
- We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.
We believe the pipeline will benefit the industry by creating better models.
---
**Distillation: Smaller Models Can Be Powerful Too**
- We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future.
- Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.
DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1.
We slightly change their configs and tokenizers. Please use our setting to run these models.
## 4. Evaluation Results
### DeepSeek-R1-Evaluation
For all our models, the maximum generation length is set to 32,768 tokens. For benchmarks requiring sampling, we use a temperature of $0.6$, a top-p value of $0.95$, and generate 64 responses per query to estimate pass@1.
**We recommend adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:**
1. Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
2. Avoid adding a system prompt; all instructions should be contained within the user prompt.
3. For mathematical problems, it is advisable to include a directive in your prompt such as: "put your final answer within \boxed{}".
4. When evaluating model performance, it is recommended to conduct multiple tests and average the results.
This code repository and the model weights are licensed under the [MIT License](https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE).
DeepSeek-R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs. Please note that:
- DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, DeepSeek-R1-Distill-Qwen-14B and DeepSeek-R1-Distill-Qwen-32B are derived from [Qwen-2.5 series](https://github.com/QwenLM/Qwen2.5), which are originally licensed under [Apache 2.0 License](https://huggingface.co/Qwen/Qwen2.5-1.5B/blob/main/LICENSE), and now finetuned with 800k samples curated with DeepSeek-R1.
- DeepSeek-R1-Distill-Llama-8B is derived from Llama3.1-8B-Base and is originally licensed under [llama3.1 license](https://huggingface.co/meta-llama/Llama-3.1-8B/blob/main/LICENSE).
- DeepSeek-R1-Distill-Llama-70B is derived from Llama3.3-70B-Instruct and is originally licensed under [llama3.3 license](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE).
## 8. Citation
```
```
## 9. Contact
If you have any questions, please raise an issue or contact us at [service@deepseek.com](service@deepseek.com).
