DeepSeek API

API only — no SDK version · active · verified Tue Mar 24

DeepSeek has NO official Python SDK. The correct integration pattern is using the openai package with base_url='https://api.deepseek.com'. The deepseek PyPI package (1.0.0) is a stub placeholder — do not use it. DeepSeek API is fully OpenAI-compatible. Two primary models: deepseek-chat (V3.2, general purpose) and deepseek-reasoner (R1, chain-of-thought reasoning).

Warnings

breaking No official DeepSeek Python SDK exists. The 'deepseek' PyPI package (1.0.0) is an empty stub with no functionality. LLMs hallucinate 'import deepseek' or 'from deepseek import Client' — neither works.
Fix: pip install openai; use OpenAI(api_key=DEEPSEEK_API_KEY, base_url='https://api.deepseek.com')
breaking R1 reasoning model name is 'deepseek-reasoner' not 'deepseek-r1'. Wrong model name returns 404 or model not found error.
Fix: model='deepseek-reasoner' for R1 thinking model. model='deepseek-chat' for V3.2 general model.
gotcha deepseek-reasoner returns an extra reasoning_content field on the message object. Standard OpenAI response parsing misses the chain-of-thought output.
Fix: Access response.choices[0].message.reasoning_content for thinking output, response.choices[0].message.content for final answer.
gotcha base_url must be 'https://api.deepseek.com' — no trailing slash, no /v1 suffix needed. The /v1 path exists for OpenAI compatibility but has no relationship to model version.
Fix: base_url='https://api.deepseek.com' — both with and without /v1 work but without is canonical.
gotcha API key env var is DEEPSEEK_API_KEY not OPENAI_API_KEY. OpenAI SDK will silently use OPENAI_API_KEY if DEEPSEEK_API_KEY is not explicitly passed.
Fix: Always pass api_key=os.environ['DEEPSEEK_API_KEY'] explicitly to OpenAI() constructor.
gotcha deepseek-reasoner does not support system prompt in the same way. System prompts are accepted but may be converted to user messages internally. Behavior differs from deepseek-chat.
Fix: For deepseek-reasoner use user messages for instructions, not system prompts.

Install

pip install openai Python (correct — use openai SDK with DeepSeek base_url)

Imports

DeepSeek via OpenAI SDK

from openai import OpenAI
import os

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

response = client.chat.completions.create(
    model='deepseek-chat',
    messages=[
        {'role': 'system', 'content': 'You are a helpful assistant'},
        {'role': 'user', 'content': 'Hello'}
    ],
    stream=False
)
print(response.choices[0].message.content)

No deepseek Python package exists. The deepseek PyPI stub (1.0.0) has no functionality. Always use openai SDK with base_url='https://api.deepseek.com'.

deepseek-reasoner (R1 thinking model)

from openai import OpenAI

client = OpenAI(
    api_key='DEEPSEEK_API_KEY',
    base_url='https://api.deepseek.com'
)

response = client.chat.completions.create(
    model='deepseek-reasoner',  # R1 thinking model
    messages=[{'role': 'user', 'content': 'Solve: 2x + 5 = 13'}]
)
# Access reasoning content separately
reasoning = response.choices[0].message.reasoning_content
answer = response.choices[0].message.content
print(reasoning)
print(answer)

Model name is 'deepseek-reasoner' not 'deepseek-r1'. Reasoning model returns extra reasoning_content field not present in standard OpenAI responses.

Quickstart

Minimal DeepSeek API call using openai SDK with base_url override.

# pip install openai
from openai import OpenAI
import os

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

response = client.chat.completions.create(
    model='deepseek-chat',
    messages=[
        {'role': 'system', 'content': 'You are a helpful assistant'},
        {'role': 'user', 'content': 'What is the capital of France?'}
    ]
)
print(response.choices[0].message.content)

view raw JSON →