DeepSeek API tutorial

How to authenticate, which endpoint to call, and the mistakes that cost people an afternoon.

The DeepSeek API is OpenAI-compatible: the official documentation shows requests built with the standard OpenAI SDKs. In practice that means three values decide whether your first call works — the API key, the base URL and the model ID.

1. Create an account and an API key

• Sign in to the developer platform: platform.deepseek.com/.
• Create an API key in the console and store it in an environment variable rather than in source code.
• Full details: API documentation

2. Point your client at the right endpoint

https://api.deepseek.com
Anthropic-format endpoint: https://api.deepseek.com/anthropic

If you are migrating existing code, keep your current SDK and change only base_url and the model name. If you see 401 responses, the key is usually wrong; if you see 404 on the model, the model ID is wrong or deprecated.

3. A minimal working request

from openai import OpenAI

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

resp = client.chat.completions.create(
    model="deepseek-flash",
    messages=[{"role": "user", "content": "Write a Python function that retries an HTTP call."}],
)
print(resp.choices[0].message.content)

4. Watch out for these

• The same model is labelled deepseek-flash in the English docs and deepseek-v4-flash in the Chinese docs.
• Two official pages describe the V4-Pro / V4.1-Flash routing differently; check the pricing page before you commit to a model ID.
• The Chinese legal entity name is not published on the official site.
• Old model IDs get retired: check the model list before hard-coding a name in production.
• Context limits are documented per model — sending more input than the window allows is the most common cause of rejection in long-document workloads.

5. Controlling cost

• Cache-hit input tokens are cheaper than cache-miss input tokens, so keep prompts stable where possible.
• Batch or off-peak options are documented for some models and are billed below the standard rate.
• Current rates: official pricing page.

6. Next steps

• DeepSeek overview: what the product is, which models exist and what each is documented to do.
• Published DeepSeek pricing: free allowances, token rates and subscription tiers, with sources.
• DeepSeek vs Western AI models: how the interfaces, context limits and availability compare.
• DeepSeek in the AI-Mind directory: ratings, categories and alternatives.

Facts on this page were checked against the official pages linked above on 2026-09-19. Prices and model IDs change frequently: confirm them on the vendor’s own pricing page before you rely on them.