SETUP GUIDE

How to run Claude Code with DeepSeek

DeepSeek is one of the most popular LLMs among developers in China; its V4 series is strong at code generation and reasoning. DeepSeek officially provides an Anthropic-compatible endpoint, making it a first-class backend for Claude Code.

Engine support (probe-verified, 2026-07-10)

EngineDeepSeek support
Claude Code✅ Supported (Anthropic-compatible endpoint)
Codex— Not yet

Step 1 — Get a DeepSeek API key

Open the DeepSeek platform → sign up / log in → “API Keys” → create a key (starts with sk-).

Console: https://platform.deepseek.com/api_keys

Step 2 — Configure the Snail launcher (3 steps)

  1. Download the launcher for Mac / Windows from justalittlesnail.ai;
  2. Paste the API key on the DeepSeek card (the launcher validates connectivity immediately);
  3. Pick a model and click Launch Claude Code — a dedicated terminal window opens.

Recommended models

Main: deepseek-v4-pro (complex tasks) · Fast: deepseek-v4-flash (subtasks / summaries). After saving the key, click “Refresh” on the card to pull the live model list.

FAQ

Why can't I use DeepSeek with Codex?
Codex CLI only supports the OpenAI Responses API protocol since 2026 (Chat Completions was removed). DeepSeek currently offers Chat Completions and an Anthropic-compatible endpoint, but no /responses endpoint — so Codex cannot connect at the protocol level. Use the Claude Code engine for DeepSeek, or pick Qwen / MiMo / MiniMax / StepFun / Baidu Qianfan for Codex.
Claude Code says “Both ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY set”?
Third-party Anthropic-compatible endpoints must use ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY. Clear the leftover variable (or upgrade the Snail launcher to v1.1.73+, which cleans the session environment automatically).

Multi-window sessions, a different model in each window

Download the Snail launcher