Connect to OpenAI (GPT-5.2 / Codex)

Estimated time: 10 minutes

OpenAI is the company behind ChatGPT, providing some of the most powerful AI models available. GPT-5.2, released in December 2025, is the current flagship model, while GPT-5.2-Codex is a version specifically optimized for coding tasks.

Note for users in China

OpenAI requires a proxy to access. Please complete 1.3 Network Configuration first.

If you can't set up the network, consider using domestic alternatives like Zhipu or DeepSeek.

---

What You'll Be Able to Do

• Log in to OpenCode directly with a ChatGPT subscription
• Or connect to OpenAI with an API Key
• Use the latest models like GPT-5.2 and GPT-5.2-Codex

---

Two Login Methods

Method	Best for	Advantages
ChatGPT Subscription (Recommended)	Existing Plus/Pro members	Use directly, no extra cost
API Key	Users without subscription	Pay-as-you-go, pay only for what you use

---

Method 1: ChatGPT Subscription Login (Recommended)

If you're already a ChatGPT Plus ($20/month) or ChatGPT Pro ($200/month) subscriber, you can log in directly with your subscription without needing to purchase API separately.

🎒 Before You Start

• [ ] Completed 1.2 Installation, able to run opencode
• [ ] Completed 1.3 Network Configuration (required for users in China)
• [ ] Have a ChatGPT Plus or Pro subscription

Follow Along

Step 1: Start OpenCode

opencode

Step 2: Connect to OpenAI

Enter:

/connect

Use arrow keys to find OpenAI, press Enter.

Step 3: Select Login Method

You'll see the login method selection:

┌ Select auth method
│
│ ChatGPT Plus/Pro (Recommended)
│ Create an API Key
│ Manually enter API Key
└

Select ChatGPT Plus/Pro, press Enter.

Step 4: Browser Authorization

OpenCode will automatically open your browser and redirect to the OpenAI login page.

1. Log in with your ChatGPT account
2. Click Authorize
3. After seeing "Authorization successful", return to the terminal

Step 5: Verify Success

Back in OpenCode, you'll see:

✓ Login successful

Now enter /models, and you should see all OpenAI models! If you've just enabled a new model but the list hasn't updated, run opencode models --refresh in your system terminal to refresh the cache, then return to /models.

---

Method 2: API Key Login

If you don't have a ChatGPT subscription, you can use the API Key method with pay-as-you-go pricing.

🎒 Before You Start

• [ ] Completed 1.2 Installation
• [ ] Completed 1.3 Network Configuration (required for users in China)
• [ ] Have an international payment method (Visa / Mastercard)

Follow Along

Step 1: Register an OpenAI Account

Visit: https://platform.openai.com

Click Sign up and follow the prompts to register and log in.

Step 2: Get Your API Key

1. After logging in, click API keys in the left menu
2. Or visit directly: https://platform.openai.com/api-keys
3. Click Create new secret key
4. Give your key a name (like opencode)
5. Copy and save immediately

Shown only once!

The API Key will only be displayed once. You won't be able to see it after closing the window. Please copy it to a safe place immediately.

Step 3: Add Credits

1. Visit: https://platform.openai.com/settings/organization/billing
2. Click Add payment method
3. Add a credit card and add credits (we recommend starting with $5 to try it out)

Step 4: Connect in OpenCode

Start OpenCode and enter:

/connect

Select OpenAI, then choose Manually enter API Key.

Paste your API Key:

┌ API key
│ sk-xxxxxxxxxxxxxxxxxxxxxxxx
└ Press Enter to confirm

---

Selecting a Model

After successfully connecting, enter:

/models

You'll see OpenAI's model list. Here are the latest model recommendations as of January 2026:

GPT-5.2 Series (Latest Flagship)

Model	Features	Best for
gpt-5.2-chat-latest (Instant)	Fast response, daily use	Q&A, documentation, translation
gpt-5.2 (Thinking)	Deep thinking, complex tasks	Analysis, planning, difficult problems
gpt-5.2-pro	Most powerful and precise	Professional work, research
gpt-5.2-codex	Code specialized	Writing code, refactoring, debugging

Which to Choose?

• Daily use: gpt-5.2-chat-latest (fast)
• Complex tasks: gpt-5.2 (powerful)
• Coding: gpt-5.2-codex (specialized)

Use arrow keys to select, press Enter to confirm.

---

GPT-5.2-Codex Deep Dive

What is GPT-5.2-Codex?

GPT-5.2-Codex is a specialized coding model released by OpenAI on December 18, 2025. It's built on GPT-5.2 with further optimization specifically for software engineering tasks.

Its features include:

• Long-running tasks: Supports context compression, can handle large codebases
• Large-scale refactoring: Excellent at code migration, refactoring, and other complex tasks
• Enhanced visual understanding: Can understand designs, screenshots, and technical diagrams
• Cybersecurity capabilities: Outstanding performance in vulnerability discovery and code auditing

When to Use Codex?

Scenario	Recommended Model	Reason
Writing complex features	✅ GPT-5.2-Codex	Good long-running task support
Large-scale refactoring	✅ GPT-5.2-Codex	Specifically optimized
Code migration	✅ GPT-5.2-Codex	Can handle large codebases
Debugging	✅ GPT-5.2-Codex	Deeper understanding of code logic
Daily Q&A	GPT-5.2 Instant	Faster
Writing documentation	GPT-5.2 Instant	General capabilities sufficient

Simply put: Use Codex for coding, use GPT-5.2 Instant for everything else.

Switch to Codex

Enter:

/models

Use arrow keys to find openai / gpt-5.2-codex, press Enter to confirm.

Try Codex Hands-on

After switching to Codex, try this prompt:

Help me write a Node.js function that:
1. Reads all .md files in a specified directory
2. Extracts the title from each file (the first line starting with #)
3. Generates a table of contents index file called index.md

You'll find that Codex is particularly clear when understanding multi-step tasks, and the generated code structure is excellent.

Codex Usage Tips

1. Provide enough context

The more Codex understands your project, the more accurate its output. Use @ to reference relevant files:

@src/routes/user.ts @src/middleware/auth.ts

Help me add permission checking to the user routes

2. Describe your goals clearly

Don't just say "optimize this function", be clear about what to optimize:

This function has performance issues and is slow with large data.
Please optimize it with these goals:
1. Reduce database query count
2. Add caching
3. Keep the existing interface unchanged

3. Let it analyze before acting

For complex tasks, you can ask Codex to analyze first:

First analyze the code structure of this module, then tell me how you plan to refactor it

After it finishes analyzing and you confirm the direction is right, then let it take action.

4. Leverage visual capabilities

Codex can understand screenshots and design files. You can:

@design.png

Implement the page based on this design

---

Checklist ✅

• [ ] Can see OpenAI models in /models
• [ ] Can send messages and receive AI responses
• [ ] Can switch to the GPT-5.2-Codex model

---

Troubleshooting

Issue	Cause	Solution
No response after browser authorization	Network issue	Check proxy, refresh and retry
connection timeout	Network issue	Check proxy configuration, see 1.3 Network Configuration
API key invalid	Key copied incorrectly	Copy again, make sure there are no extra spaces
insufficient_quota	Insufficient balance	Add credits in OpenAI console
Page won't load	Proxy not on or unstable	Try a different proxy node

---

Pricing

ChatGPT Subscription Users

• Plus: $20/month, includes all GPT-5.2 series models
• Pro: $200/month, higher usage limits, access to GPT-5.2 Pro

API Users

Pay per token (1 token ≈ 0.75 English words ≈ 0.5 Chinese characters):

Model	Input Price	Output Price
gpt-5.2 / gpt-5.2-chat-latest	$1.75/million	$14/million
gpt-5.2-pro	$21/million	$168/million

Typical daily usage is around $0.1-0.5.

---

Next Steps

• Return to 1.4 Overview to choose other models
• Or proceed to 2.1 Interface & Basic Operations to learn more

Having trouble?

Stuck during configuration? Join the community to connect with 2000+ fellow learners and get real-time help.