# DeepSeek-R1 — Open-Weight Reasoning Model Rivaling OpenAI o1

> DeepSeek-R1 is the open-weight reasoning model that matches OpenAI o1 on math, code, science benchmarks. Streaming chain-of-thought visible. MIT-licensed.

## Install

Copy the content below into your project:

## Quick Use

1. Hosted: same DeepSeek API key, set `model="deepseek-reasoner"`
2. Local: `ollama pull deepseek-r1:7b && ollama run deepseek-r1:7b`
3. Print `response.message.reasoning_content` to see the full chain-of-thought

---

## Intro

DeepSeek-R1 is the open-weight reasoning model that achieves o1-level performance on AIME / MATH / GPQA / Codeforces while shipping its full chain-of-thought to the user. Distilled smaller versions (1.5B, 7B, 32B, 70B) make local reasoning practical on consumer hardware. MIT license, full weights public. Best for: hard reasoning tasks (math, science, complex code) where you need a reasoning model but want open weights. Works with: DeepSeek API, Ollama (distilled), vLLM, llama.cpp. Setup time: 2 minutes.

---

### Hosted API

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com/v1",
    api_key=os.environ["DEEPSEEK_API_KEY"],
)

response = client.chat.completions.create(
    model="deepseek-reasoner",  # R1
    messages=[{"role": "user", "content":
        "Prove that the square root of 2 is irrational"}],
)

# R1 streams reasoning + final answer
for choice in response.choices:
    print("REASONING:", choice.message.reasoning_content)
    print("ANSWER:", choice.message.content)
```

Unlike o1, R1's reasoning is visible — useful for debugging, education, and trust.

### Local via Ollama (distilled)

```bash
ollama pull deepseek-r1:1.5b   # ~1GB, runs on a laptop
ollama pull deepseek-r1:7b     # ~5GB
ollama pull deepseek-r1:14b    # ~9GB
ollama pull deepseek-r1:32b    # ~20GB, M2 Max territory
ollama pull deepseek-r1:70b    # ~40GB, beefy server
```

The 7B distillation often outperforms GPT-4o on competition math while being free and fast on a single 4090.

### When to use R1 vs V3

| Task | Pick |
|---|---|
| Math proofs, competition problems | R1 |
| Step-by-step debugging | R1 |
| Quick chitchat, summaries | V3 (cheaper, faster) |
| Tool-use heavy agent | V3 (R1's tool support is weaker) |
| Need visible CoT for audit | R1 |

### Pricing

| Source | Input $/1M tok | Output $/1M tok |
|---|---|---|
| DeepSeek API | $0.55 | $2.19 |
| OpenAI o1 (compare) | $15.00 | $60.00 |
| OpenAI o1-mini (compare) | $3.00 | $12.00 |
| Local distilled | $0 | $0 |

---

### FAQ

**Q: Why does R1 show its reasoning when o1 hides it?**
A: DeepSeek published the full RL training methodology. Visible CoT is part of the value proposition — auditability, debugging, education. OpenAI considers o1's CoT a competitive moat.

**Q: How much slower is R1 vs V3?**
A: R1 spends extra tokens on reasoning before the final answer — typically 3-10× more output tokens, so 3-10× slower wall-clock latency on equal infra. The cost difference reflects this.

**Q: Are the distilled R1 versions trained from scratch?**
A: No — they're knowledge-distilled from full R1 into Llama / Qwen base models. The 7B distill is `Llama-3.1-8B + R1 distillation`, the 32B is `Qwen-2.5-32B + R1 distillation`, etc. Performance trades off with base.

---

## Source & Thanks

> Built by [DeepSeek](https://github.com/deepseek-ai). Weights MIT-licensed.
>
> [deepseek-ai/DeepSeek-R1](https://github.com/deepseek-ai/DeepSeek-R1) — ⭐ 90,000+

---

<!-- ZH -->

## 快速使用

1. 托管：用同一个 DeepSeek API key，设 `model="deepseek-reasoner"`
2. 本地：`ollama pull deepseek-r1:7b && ollama run deepseek-r1:7b`
3. 打印 `response.message.reasoning_content` 看完整 chain-of-thought

---

## 简介

DeepSeek-R1 是开源权重的推理模型，在 AIME / MATH / GPQA / Codeforces 上达到 o1 级别，同时把完整 chain-of-thought 推给用户。蒸馏的小版本（1.5B / 7B / 32B / 70B）让消费级硬件上跑本地推理变得现实。MIT 许可证，完整权重公开。适合需要推理模型又要开源权重的硬推理任务（数学、科学、复杂代码）。兼容 DeepSeek API / Ollama（蒸馏）/ vLLM / llama.cpp。装机时间 2 分钟。

---

### 托管 API

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com/v1",
    api_key=os.environ["DEEPSEEK_API_KEY"],
)

response = client.chat.completions.create(
    model="deepseek-reasoner",  # R1
    messages=[{"role": "user", "content":
        "Prove that the square root of 2 is irrational"}],
)

# R1 流式输出推理 + 最终答案
for choice in response.choices:
    print("REASONING:", choice.message.reasoning_content)
    print("ANSWER:", choice.message.content)
```

不像 o1，R1 的推理过程可见 —— 对调试、教育、信任有用。

### 本地 Ollama（蒸馏）

```bash
ollama pull deepseek-r1:1.5b   # ~1GB，笔记本上能跑
ollama pull deepseek-r1:7b     # ~5GB
ollama pull deepseek-r1:14b    # ~9GB
ollama pull deepseek-r1:32b    # ~20GB，M2 Max 级别
ollama pull deepseek-r1:70b    # ~40GB，强服务器
```

7B 蒸馏版在竞赛数学上常常胜过 GPT-4o，单张 4090 上免费且快。

### R1 vs V3 怎么选

| 任务 | 选 |
|---|---|
| 数学证明、竞赛题 | R1 |
| 逐步调试 | R1 |
| 快速闲聊、摘要 | V3（更便宜、更快） |
| 工具使用重的 agent | V3（R1 工具支持较弱） |
| 需要可审计 CoT | R1 |

### 价格

| 来源 | 输入 $/1M tok | 输出 $/1M tok |
|---|---|---|
| DeepSeek API | $0.55 | $2.19 |
| OpenAI o1（对比） | $15.00 | $60.00 |
| OpenAI o1-mini（对比） | $3.00 | $12.00 |
| 本地蒸馏 | $0 | $0 |

---

### FAQ

**Q: 为啥 R1 显示推理过程而 o1 隐藏？**
A: DeepSeek 公开了完整 RL 训练方法。可见 CoT 是价值主张的一部分 —— 可审计、可调试、可教育。OpenAI 把 o1 的 CoT 当成竞争壁垒。

**Q: R1 比 V3 慢多少？**
A: R1 在最终答案前多花 token 推理 —— 通常输出 token 多 3-10 倍，同等基础设施下延迟 3-10 倍。价格差反映了这个。

**Q: R1 蒸馏版本是从零训练的吗？**
A: 不是 —— 从完整 R1 知识蒸馏进 Llama / Qwen 基础模型。7B 蒸馏是 `Llama-3.1-8B + R1 distillation`，32B 是 `Qwen-2.5-32B + R1 distillation` 等。性能跟基础模型有取舍。

---

## 来源与感谢

> Built by [DeepSeek](https://github.com/deepseek-ai). Weights MIT-licensed.
>
> [deepseek-ai/DeepSeek-R1](https://github.com/deepseek-ai/DeepSeek-R1) — ⭐ 90,000+


---
Source: https://tokrepo.com/en/workflows/deepseek-r1-open-weight-reasoning-model-rivaling-openai-o1
Author: DeepSeek