DeepSeek API

V4-Flash and V4-Pro — DeepSeek's frontier models with 1M token context, native thinking mode, and full OpenAI / Anthropic API compatibility. Drop-in replacement for your existing stack.

Pricing

Billed per 1M tokens in USD. Cost = tokens × price, deducted from topped-up balance first, then granted balance. Cache-hit input tokens are 50× cheaper than cache-miss — use context caching to cut costs dramatically.

Model Comparison

Key Features

Thinking Mode

Both V4-Flash and V4-Pro support native thinking/reasoning. Enable it via the thinking parameter and control effort with reasoning_effort:

{
  "model": "deepseek-v4-flash",
  "thinking": { "type": "enabled" },
  "reasoning_effort": "medium",
  "messages": [{"role": "user", "content": "Explain quantum entanglement simply"}]
}

API Endpoints

DeepSeek supports both OpenAI and Anthropic API formats — same API key works for both.

Base URLs

Code Examples

Authentication

Get your API key from platform.deepseek.com. All requests require a Bearer token:

Authorization: Bearer <DEEPSEEK_API_KEY>

Chat Completion

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hello!"}
    ],
    "stream": false
  }'

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_DEEPSEEK_API_KEY",
    base_url="https://api.deepseek.com"
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"}
    ],
    stream=False
)
print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
    apiKey: process.env.DEEPSEEK_API_KEY,
    baseURL: "https://api.deepseek.com"
});

const response = await client.chat.completions.create({
    model: "deepseek-v4-flash",
    messages: [
        { role: "system", content: "You are a helpful assistant." },
        { role: "user", content: "Hello!" }
    ],
    stream: false
});
console.log(response.choices[0].message.content);

Streaming

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Tell me a story"}],
    "stream": true
  }'

Responses return as Server-Sent Events (SSE) with data: {...} lines and a final [DONE] sentinel.

Context Caching

Repeat the same prefix across requests to trigger automatic caching. Cache-hit tokens billed at the discounted rate — no code changes required beyond reusing the prefix.

# First request — full price (cache miss)
curl https://api.deepseek.com/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {"role": "system", "content": "You are a coding assistant. Here is the project context: [long docs]..."},
      {"role": "user", "content": "Now help me with this specific question..."}
    ]
  }'

# Subsequent requests — prefix cached at $0.0028/1M
curl https://api.deepseek.com/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {"role": "system", "content": "You are a coding assistant. Here is the project context: [long docs]..."},
      {"role": "user", "content": "A different question with same prefix..."}
    ]
  }'

Token Usage

Tokens are the basic billing units — roughly 1 English character ≈ 0.3 tokens, 1 Chinese character ≈ 0.6 tokens. Actual counts vary by model; the definitive count comes from the usage field in the API response.

Token Cost Estimator

Rough estimate based on average token ratios. Actual counts come from the model's usage field.

Rate Limits

Concurrency limits are enforced at the account level regardless of which API key is used. A request counts as concurrent from send until response completes.

Error Codes

Agent Integrations

DeepSeek integrates directly with popular coding agents. Configure via OpenAI-compatible or Anthropic-compatible format depending on your tool.

FAQ

Full FAQ →

Getting Started

Browse the rest of the Dump collection.