Source: catalog-fallback

DeepSeek API documentation, organized for your first call

Start with the DeepSeek quick answers below, then browse the curated official catalog. The page also calls/api/documentsin the browser to refresh page titles and descriptions from each official docs site. If a crawl fails, it falls back to the built-in official link summary.

Providers

Pages

Updated

2026年7月19日 18:10

DeepSeek API quick answers

The five things developers ask before their first call. Each answer links to a full guide.

First API call

What is the DeepSeek API base URL?

DeepSeek is OpenAI-compatible. Keep the standard openai library and change only base_url, api_key, and the model name.

base_url="https://api.deepseek.com/v1"Python quickstart guide

Model selection

Which DeepSeek model ID should I use?

Use deepseek-v4-flash as the default for chat, streaming, and repeated tool steps. Switch to deepseek-v4-pro for harder reasoning or review-heavy work. Both have a 1M-token context window.

Pro vs Flash guide

Thinking & tool calls

How do thinking mode and tool calls work?

Thinking mode returns a reasoning_content field alongside the answer, and DeepSeek supports OpenAI-style function calling / tools. Use Flash for high-frequency tool loops and Pro for harder chains.

Thinking & tool-call guide

Pricing & rate limits

What does the DeepSeek API cost?

V4 Pro: $0.435 input / $0.87 output per 1M tokens ($0.003625 cache-hit input). V4 Flash: $0.14 input / $0.28 output per 1M tokens ($0.0028 cache-hit input). Rate limits are account-level and isolated per user_id.

Rate limit & user_id guide

Troubleshooting

OpenAI-compatible or Anthropic-compatible endpoint?

Use the /v1 OpenAI-compatible endpoint for standard SDKs, or DeepSeek's Anthropic-compatible endpoint for Claude Code. Always call the real model IDs (deepseek-v4-pro / deepseek-v4-flash); legacy aliases are being retired.

OpenAI vs Anthropic routing guide