How to use DeepSeek models in Claude Code

How to set up Claude Code with DeepSeek

1. Get a DeepSeek API key

   Create one on platform.deepseek.com under API keys, and add balance to the account; usage is billed per token.

2. Install Claude Code if you haven't

   DeepSeek's guide installs it with npm: Node.js 18 or later, Git for Windows on Windows, then npm install -g @anthropic-ai/claude-code and claude --version to check.

3. Set the environment variables (macOS and Linux)

   These point Claude Code at DeepSeek's Anthropic-compatible endpoint, use deepseek-flash with its 1M context for the main model and subagents, and raise the effort level and auto-compact window. Put them in your shell profile to keep them.

   export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
   export ANTHROPIC_AUTH_TOKEN=<your DeepSeek API Key>
   export ANTHROPIC_MODEL=deepseek-flash[1m]
   export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-flash[1m]
   export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-flash[1m]
   export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-flash
   export CLAUDE_CODE_SUBAGENT_MODEL=deepseek-flash
   export CLAUDE_CODE_EFFORT_LEVEL=max
   export CLAUDE_CODE_AUTO_COMPACT_WINDOW=786432

4. Or on Windows (PowerShell)

   The same settings as PowerShell environment variables for the current session.

   $env:ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
   $env:ANTHROPIC_AUTH_TOKEN="<your DeepSeek API Key>"
   $env:ANTHROPIC_MODEL="deepseek-flash[1m]"
   $env:ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek-flash[1m]"
   $env:ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek-flash[1m]"
   $env:ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-flash"
   $env:CLAUDE_CODE_SUBAGENT_MODEL="deepseek-flash"
   $env:CLAUDE_CODE_EFFORT_LEVEL="max"
   $env:CLAUDE_CODE_AUTO_COMPACT_WINDOW="786432"

5. Start Claude Code in your project

   Run cd into the project and claude. Everything else, such as CLAUDE.md, skills, hooks and MCP servers, is configured as usual.

Source: DeepSeek API docs, Integrate with Claude Code

How Claude model names map to DeepSeek models

When Claude Code (or the Claude desktop app) sends a Claude model name to DeepSeek's endpoint, DeepSeek maps it like this. That's why the setup above sets every model variable to deepseek-flash: otherwise anything that asks for Opus is billed at V4 Pro prices.

What it costs

DeepSeek API prices in US dollars per million tokens on November 6, 2026. Off-peak rates (outside 01:00–04:00 and 06:00–10:00 UTC on weekdays) are half of peak. Web search in Claude Code also works against DeepSeek's API, but each search adds model tokens to summarize results and is billed under DeepSeek's Search API.

Using DeepSeek in Codex

DeepSeek also supports Codex, through the Responses API. Its guide offers a setup script that backs up ~/.codex/config.toml, writes a ~/.codex/models.json catalog with deepseek-flash and deepseek-v4-pro, and adds a DeepSeek provider; one configuration then covers the Codex CLI, the ChatGPT desktop app and the VS Code extension. Read the script before running it, or follow the manual steps in the same guide. Built-in web search is turned off by the script and is available only with deepseek-v4-pro. For Codex's own instruction files, see the Codex AGENTS.md guide.

Claude Code with DeepSeek vs DeepSeek Harness

• Claude Code with DeepSeek: you keep Claude Code's interface and habits, including CLAUDE.md, skills, hooks and plugins, and pay DeepSeek's per-token prices instead of a Claude plan.
• DeepSeek Harness: DeepSeek's own open-source agent app, built around plugins, with a desktop app and web UI; it can also run Claude Code or Codex as subagents. See the DeepSeek Harness guide.
• Either way, the models are the same deepseek-flash and deepseek-v4-pro, billed by DeepSeek.
• To switch back to Anthropic's models, unset the variables (or open a new shell without them) and sign in to Claude Code as usual.

Cheap models, working all day

Running Claude Code or DeepSeek Harness on DeepSeek's prices makes it affordable to keep agents busy. Crewmark turns those sessions into AI employees with a role, a goal and working hours, and wakes each one on a heartbeat to take the next issue from a board.

1. Use the setup you already have

   An employee runs in a Claude Code, Codex, DeepSeek Harness or OpenCode session on your machine, with your environment and keys.

2. Let the heartbeat start the work

   During working hours it wakes every 15, 30 or 60 minutes in the same conversation and takes the next issue; see how to make Claude Code proactive.

3. Keep the bill in check

   Working hours and one issue at a time keep usage predictable; see how to reduce Claude Code token usage.

Claude Code with DeepSeek FAQ