Codex: Codex with DeepSeek
Last updated: 2026-08-31
DeepSeek is a leading large language model with excellent performance in code generation. This lesson covers how to use DeepSeek models within Codex.
📋 Prerequisites: Codex CLI installed, basic API Key configuration knowledge
1. What You Will Learn
- Why use DeepSeek with Codex
- Official integration method
- CCSwitch third-party tool configuration
- DeepSeek vs OpenAI model comparison
2. Why Use DeepSeek
| Advantage | Description |
|---|---|
| Lower cost | DeepSeek API pricing is significantly lower than OpenAI |
| Direct access | No proxy needed for users in China |
| Strong coding | DeepSeek Coder excels at code tasks |
| Better Chinese | High-quality Chinese understanding and generation |
(1) Price Comparison
| Model | Input Price | Output Price |
|---|---|---|
| GPT-5 Codex | $2.50/1M tokens | $10.00/1M tokens |
| DeepSeek V3 | ¥1.00/1M tokens | ¥2.00/1M tokens |
| DeepSeek Coder | ¥1.00/1M tokens | ¥2.00/1M tokens |
3. Official Integration Method
Codex CLI supports DeepSeek via the OpenAI-compatible API.
(1) Get a DeepSeek API Key
TEXT
📖 Display only
1. Visit platform.deepseek.com
2. Register / log in
3. Go to the API Keys page
4. Create a new API Key
5. Copy and save (shown only once)
(2) Configure Environment Variables
BASH
# macOS/Linux
export OPENAI_API_KEY="sk-your-deepseek-key"
export OPENAI_BASE_URL="https://api.deepseek.com"
# Windows PowerShell
$env:OPENAI_API_KEY="sk-your-deepseek-key"
$env:OPENAI_BASE_URL="https://api.deepseek.com"
# Permanent (add to ~/.zshrc or ~/.bashrc)
echo 'export OPENAI_API_KEY="sk-your-deepseek-key"' >> ~/.zshrc
echo 'export OPENAI_BASE_URL="https://api.deepseek.com"' >> ~/.zshrc
source ~/.zshrc
(3) Launch Codex
BASH
# Launch with DeepSeek model
codex --model deepseek-coder
# Or use DeepSeek V3
codex --model deepseek-chat
▶ Example 1: Alice Switches to DeepSeek
BASH
# Alice wants to save costs, switches to DeepSeek
export OPENAI_API_KEY="sk-deepseek-xxx"
export OPENAI_BASE_URL="https://api.deepseek.com"
cd my-project
codex --model deepseek-coder
# Alice enters:
> "Add OAuth2 support to the auth module"
# DeepSeek Coder executes the task normally
4. CCSwitch Tool
CCSwitch is a community-developed third-party tool for quickly switching between different model providers.
(1) Install CCSwitch
BASH
npm install -g ccswitch
(2) Configure Model Providers
BASH
# Add DeepSeek config
ccswitch add deepseek \
--api-key "sk-your-deepseek-key" \
--base-url "https://api.deepseek.com" \
--model "deepseek-coder"
# Add OpenAI config
ccswitch add openai \
--api-key "sk-your-openai-key" \
--base-url "https://api.openai.com/v1" \
--model "gpt-5-codex"
(3) Quick Switch
BASH
# Switch to DeepSeek
ccswitch use deepseek
# Switch to OpenAI
ccswitch use openai
# View current config
ccswitch current
▶ Example 2: Bob's Multi-Model Workflow
BASH
# Daily dev with DeepSeek (cost-saving)
ccswitch use deepseek
codex "Fix the CSS issue on the login page"
# Complex architecture tasks with OpenAI (more capable)
ccswitch use openai
codex "Design a microservice decomposition plan, generate architecture docs"
# Code review with DeepSeek Coder
ccswitch use deepseek
codex "Review the code quality of PR #15"
5. DeepSeek vs OpenAI Model Comparison
| Dimension | DeepSeek Coder | GPT-5 Codex |
|---|---|---|
| Code generation | Excellent | Excellent |
| Complex reasoning | Good | Stronger |
| Chinese understanding | Better | Good |
| Cost | Low | High |
| Response speed | Fast | Fast |
| Multi-file refactoring | Good | Stronger |
6. auth.json Configuration
You can also manage multiple API Keys via a configuration file:
JSON
{
"OPENAI_API_KEY": "sk-your-deepseek-key",
"OPENAI_BASE_URL": "https://api.deepseek.com",
"model": "deepseek-coder"
}
BASH
# Write config
mkdir -p ~/.codex
cat > ~/.codex/auth.json << 'EOF'
{
"OPENAI_API_KEY": "sk-deepseek-xxx",
"OPENAI_BASE_URL": "https://api.deepseek.com"
}
EOF
7. Important Notes
| Note | Description |
|---|---|
| API compatibility | DeepSeek API is OpenAI-format compatible, but some advanced features may not be supported |
| Model names | Use deepseek-coder or deepseek-chat, not gpt-* |
| Base URL | Must set OPENAI_BASE_URL to point to the DeepSeek endpoint |
| Switching back to OpenAI | Remember to clear the OPENAI_BASE_URL environment variable |
| Quota | DeepSeek offers free credits for new users — check your balance |
❓ FAQ
Q Can DeepSeek fully replace OpenAI models?
A For daily development tasks, mostly yes. But for complex architectural reasoning and multi-file global refactoring, OpenAI models are still stronger. Choose based on task complexity.
Q Is CCSwitch safe?
A CCSwitch is an open-source community tool; API Keys are stored in local config files. Review the source code to confirm security.
Q Can I use both models simultaneously?
A Not within the same session. You need to end the current session, switch configuration, and restart.
Q Is DeepSeek's context window large enough?
A DeepSeek V3 supports 128K context, comparable to OpenAI models — sufficient for most projects.
Q Does Codex behave differently after switching models?
A Basic operations remain the same, but different models have different reasoning capabilities and output styles. DeepSeek has better Chinese understanding; OpenAI has stronger complex reasoning.
📖 Summary
- Connect DeepSeek via the
OPENAI_BASE_URLenvironment variable - CCSwitch enables quick switching between multiple models
- DeepSeek: lower cost, better Chinese — ideal for daily development
- Complex architecture tasks still benefit from OpenAI models
- Switching models requires restarting the Codex session
📝 Exercises
- Basic (⭐): Configure the DeepSeek API and successfully use it in Codex CLI.
- Intermediate (⭐⭐): Configure two model providers with CCSwitch and compare their performance on the same task.
- Advanced (⭐⭐⭐): Design an auto-switching strategy — simple tasks use DeepSeek, complex tasks use OpenAI — and write a switching script.
