How to beginner · 3 min read

How to extract reasoning steps from DeepSeek R1

Q: How to extract reasoning steps from DeepSeek R1

Use the deepseek-reasoner model via the OpenAI-compatible OpenAI Python SDK by sending a prompt that requests step-by-step reasoning. Parse the response's choices[0].message.content to extract the reasoning steps explicitly provided by the model.

Quick answer

Use the deepseek-reasoner model via the OpenAI-compatible OpenAI Python SDK by sending a prompt that requests step-by-step reasoning. Parse the response's choices[0].message.content to extract the reasoning steps explicitly provided by the model.

PREREQUISITES

Python 3.8+
DeepSeek API key
pip install openai>=1.0

Setup

Install the openai Python package and set your DeepSeek API key as an environment variable.

Run pip install openai to install the SDK.
Export your API key in your shell: export DEEPSEEK_API_KEY='your_api_key_here'.

bash

pip install openai

Step by step

Use the deepseek-reasoner model to request reasoning steps by prompting it explicitly. The example below sends a question and asks the model to provide detailed reasoning steps in the response.

python

import os
from openai import OpenAI

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

prompt = (
    "Explain the reasoning steps for the following problem:\n"
    "If a train travels 60 miles in 1.5 hours, what is its average speed?"
    " Please provide step-by-step reasoning."
)

response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=[{"role": "user", "content": prompt}]
)

reasoning_steps = response.choices[0].message.content
print("Reasoning steps from DeepSeek R1:\n", reasoning_steps)

output

Reasoning steps from DeepSeek R1:
 1. Identify the total distance traveled: 60 miles.
 2. Identify the total time taken: 1.5 hours.
 3. Use the formula for average speed: speed = distance / time.
 4. Calculate speed = 60 miles / 1.5 hours = 40 miles per hour.
 5. The average speed of the train is 40 mph.

Common variations

You can customize the request by adjusting max_tokens to control response length or use async calls for concurrency. You may also switch to other DeepSeek models for different tasks.

python

import asyncio
import os
from openai import OpenAI

async def async_reasoning():
    client = OpenAI(api_key=os.environ["DEEPSEEK_API_KEY"], base_url="https://api.deepseek.com")
    prompt = (
        "Explain the reasoning steps for this math problem:\n"
        "What is the sum of the first 10 natural numbers?"
        " Provide detailed step-by-step reasoning."
    )
    response = await client.chat.completions.acreate(
        model="deepseek-reasoner",
        messages=[{"role": "user", "content": prompt}],
        max_tokens=512
    )
    print("Async reasoning steps:\n", response.choices[0].message.content)

asyncio.run(async_reasoning())

output

Async reasoning steps:
 1. The first 10 natural numbers are 1, 2, 3, ..., 10.
 2. Use the formula for the sum of the first n natural numbers: n(n+1)/2.
 3. Substitute n=10: 10 * 11 / 2 = 55.
 4. The sum is 55.

Troubleshooting

If you receive incomplete reasoning, increase max_tokens to allow longer responses.
If the API returns authentication errors, verify your DEEPSEEK_API_KEY environment variable is set correctly.
For network issues, check your internet connection and DeepSeek service status.

✅

Key Takeaways

Use the deepseek-reasoner model with explicit prompts requesting step-by-step reasoning.
Parse response.choices[0].message.content to extract detailed reasoning steps.
Adjust max_tokens to control the length and completeness of reasoning outputs.

Verified 2026-04 · deepseek-reasoner

Verify ↗