DeepSeek V4 API User Guide

1. Registration & Authentication

The DeepSeek V4 API is compatible with the OpenAI SDK, which significantly reduces migration costs and allows developers to seamlessly integrate with existing AI ecosystem tools.

Obtaining an API Key

1. Visit DeepSeek Platform.
2. Click "API Keys" -> "Create new secret key" in the top right corner.
3. Note: The key is only displayed once upon creation, so save it securely.

2. Quick Start (Python Example)

First, install the official OpenAI SDK (yes, you read that right, just use the OpenAI SDK):

pip install openai

Then, simply modify two parameters: base_url and api_key.

from openai import OpenAI

client = OpenAI(
    api_key="sk-xxxxxxx",  # Your DeepSeek API Key
    base_url="https://api.deepseek.com"  # Key point!
)

response = client.chat.completions.create(
    model="deepseek-v4",  # Model name, may differ after V4 release
    messages=[
        {"role": "system", "content": "You are a helpful AI assistant."},
        {"role": "user", "content": "Hello, please introduce DeepSeek V4."},
    ],
    stream=False
)

print(response.choices[0].message.content)

3. Advanced Features

3.1 Streaming Output

To enhance user experience, it's recommended to enable streaming output, allowing users to see a typewriter-like generation effect.

stream = client.chat.completions.create(
    model="deepseek-v4",
    messages=[{"role": "user", "content": "Write a long poem about whales"}],
    stream=True,
)

for chunk in stream:
    if chunk.choices[0].delta.content is not None:
        print(chunk.choices[0].delta.content, end="")

3.2 Function Calling

DeepSeek V4 supports powerful Function Calling, capable of precisely parsing JSON parameters, making it ideal for building Agents.

3.3 JSON Mode

If you need the model to output valid JSON format, be sure to specify json_object in response_format and explicitly request JSON output in the prompt.

4. Pricing & Cost Optimization

DeepSeek V4 pricing is extremely affordable (specific prices pending official announcement, reference V3 at $0.14/M tokens). To save even more money:

• Prompt Caching: V4 introduces a context caching mechanism. If your System Prompt is long (e.g., knowledge base), subsequent requests will cost significantly less due to cache hits.
• Streamline Context: Don't stuff irrelevant chat history into it.

5. Best Practices

• Error Handling: Make sure to implement try-except, especially for rate_limit_exceeded (429) errors, with an exponential backoff retry strategy.
• Timeout Settings: Complex reasoning tasks may require longer generation times, so set timeout to 60s or more.
• System Prompt: Give V4 a clear "persona" and it will perform better.

Having issues? Check our API Documentation or join the Discord community for help.