How to beginner to intermediate · 3 min read

How to use reasoning models for math problems

Q: How to use reasoning models for math problems

Use specialized reasoning models such as DeepSeek-R1 or claude-sonnet-4-5 to solve math problems by prompting them with clear step-by-step instructions. These models excel at logical reasoning and multi-step calculations, making them ideal for math tasks.

Quick answer

Use specialized reasoning models such as DeepSeek-R1 or claude-sonnet-4-5 to solve math problems by prompting them with clear step-by-step instructions. These models excel at logical reasoning and multi-step calculations, making them ideal for math tasks.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the openai Python SDK and set your API key as an environment variable to access reasoning models like DeepSeek-R1.

bash

pip install openai>=1.0

Step by step

Use the DeepSeek-R1 model with a prompt that instructs the model to solve a math problem step-by-step. This example shows how to query the model and print the answer.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

math_problem = "Calculate the sum of the first 10 positive integers."
prompt = f"Solve this math problem step-by-step:\n{math_problem}\nStep 1:"

response = client.chat.completions.create(
    model="deepseek-reasoner",
    messages=[{"role": "user", "content": prompt}],
    max_tokens=256
)

print("Model response:\n", response.choices[0].message.content)

output

Model response:
Step 1: List the first 10 positive integers: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.
Step 2: Calculate their sum: 1+2+3+4+5+6+7+8+9+10 = 55.
Answer: 55.

Common variations

You can use other reasoning models like claude-sonnet-4-5 for math problems by changing the model parameter. Async calls or streaming responses are also possible depending on the SDK.

python

import os
import asyncio
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

async def solve_math_async():
    response = await client.chat.completions.acreate(
        model="claude-sonnet-4-5",
        messages=[{"role": "user", "content": "What is 15 multiplied by 12? Show steps."}],
        max_tokens=128
    )
    print("Async model response:\n", response.choices[0].message.content)

asyncio.run(solve_math_async())

output

Async model response:
Step 1: Multiply 15 by 10 = 150.
Step 2: Multiply 15 by 2 = 30.
Step 3: Add 150 + 30 = 180.
Answer: 180.

Troubleshooting

If the model returns incomplete answers, increase max_tokens to allow longer reasoning chains.
If the output is off-topic, clarify the prompt with explicit instructions like "Show your work step-by-step."
For rate limits, ensure your API key has sufficient quota or retry after some time.

✅

Key Takeaways

Use reasoning-specialized models like DeepSeek-R1 for complex math problems.
Prompt models with explicit step-by-step instructions to improve answer quality.
Adjust max_tokens and model choice based on problem complexity and response length.
Async and streaming calls are supported for responsive applications.
Clear prompts reduce hallucinations and improve math reasoning accuracy.

Verified 2026-04 · deepseek-reasoner, claude-sonnet-4-5

Verify ↗