DeepSeek API Key: How to Get One and What It Costs

This guide covers creating a DeepSeek API key, balance-based billing, peak-hour rates and cache reads. The cost examples use stated token assumptions; they are not predictions of your bill.

Last updated: November 10, 2026 · verified against DeepSeek's official pricing page ↗

Step 1 — Create the key

1. Go to platform.deepseek.com and sign up (email or Google account).
2. Open API keys in the left sidebar.
3. Click Create new API key, give it a name, and copy it immediately — DeepSeek shows the full key only once. If you lose it, delete the key and create a new one.
4. Store it in an environment variable, never in code or a git repo:

export DEEPSEEK_API_KEY="sk-..."

Step 2 — Add credits (prepaid billing)

DeepSeek deducts usage charges from your topped-up or granted balance. Granted balance is used first when both are available. Check current top-up requirements and any account-specific credits in the official dashboard before paying.

Check your remaining balance and available billing controls in the platform dashboard before using the API in production.

Step 3 — Understand the pricing before you scale

The model you get is deepseek-flash (DeepSeek-V4.1-Flash). Prices below are off-peak rates per 1M tokens, verified November 2026:

Two things surprise new users:

• Peak hours cost 2×. Peak windows are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday (excluding Chinese public holidays). If half your tokens run in peak windows, your blended bill is about 1.5× the off-peak estimate — see the worked examples.
• Cache hits are ~2% of the normal input price. Repeated system prompts and long static context bill at $0.003/1M instead of $0.15/1M. Design prompts so the static part stays at the front, and measure actual cache hits in your usage data. A repeated prompt does not guarantee a cache hit.

Step 4 — Make your first call

The API speaks the OpenAI schema, so most existing code needs only a base-URL change:

import os
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": "Explain cache-hit pricing in one line."}],
)
print(resp.choices[0].message.content)

The same key also works with third-party apps that accept an OpenAI-compatible endpoint and key — chat frontends, IDE extensions, and agent frameworks. The pattern is always the same: paste the key, set the base URL to https://api.deepseek.com, and select deepseek-flash as the model. Check the app's own docs for where those fields live.

What your first month will cost

Using a common chat workload — 1,000 input + 500 output tokens per request, all off-peak, no cache hits:

At 100,000 requests per month with the same token counts, Claude Opus 5.5 costs $1,400 at its listed standard rates, excluding other fees. Compare any workload yourself in the calculator — set the cache ratio to match your prompt reuse.

Key hygiene (the part everyone skips)

• One key per environment. Separate keys for dev, staging and production so a leak can be rotated without downtime.
• Never commit keys. Use environment variables or a secrets manager. If a key lands in a public repo, delete it immediately — anyone can find it within minutes and spend your balance.
• Revoke exposed keys promptly. Rotate keys according to your security requirements, and update dependent services.

FAQ

• Do I need a Chinese phone number? Check the official signup flow for requirements in your region; signup methods and eligibility can change.
• Can I pay with a credit card? Top-up payment methods vary by region; check the billing page after signup. Use the methods offered for your account; this guide does not verify payment availability.
• Is deepseek-reasoner still a valid model name? Use deepseek-flash for the Flash model in this example. Check the official model list before migrating any legacy API name.
• What's the context window? 1M tokens, with up to 384K max output.

Estimate your own monthly cost → · DeepSeek pricing deep-dive · Cheapest LLMs for coding