How to Use GPT-5 Codex with Cursor AI (2026)

OpenAI's Codex, powered by GPT-5, is one of the most capable coding models available in 2026. When paired with Cursor AI's agentic editor, it creates a powerful development environment for code generation, debugging, and refactoring.

This guide shows you how to set up and use GPT-5 Codex within Cursor AI, including configuration options, best practices, and comparisons with other available models.

What Is GPT-5 Codex?

GPT-5 Codex is OpenAI's dedicated coding model built on the GPT-5 architecture. It is optimized for:

Code generation across 50+ programming languages
Multi-file understanding and editing
Long-context reasoning (up to 200K tokens)
Executing multi-step coding plans
Understanding and generating tests, documentation, and deployment configs

Feature	GPT-5 Codex	GPT-4o	Claude Sonnet 4
Context window	200K tokens	128K tokens	200K tokens
Code optimization	Specialized	General	General
Multi-file editing	Excellent	Good	Excellent
Speed	Fast	Fast	Fast
Cost (approx)	$5/M input, $15/M output	$2.5/M in, $10/M out	$3/M in, $15/M out

Prerequisites

Before you begin, you need:

Cursor AI installed (download from cursor.com)
Cursor Pro subscription ($20/month) for premium model access, or an OpenAI API key
An OpenAI API key if you want to use your own key instead of Cursor's built-in allocation

Method 1: Use GPT-5 Codex via Cursor's Built-in Models

Cursor Pro includes access to GPT-5 Codex as part of its premium model lineup. This is the easiest approach.

Step 1: Open Model Settings

Open Cursor
Press Cmd+Shift+P (macOS) or Ctrl+Shift+P (Windows/Linux)
Type "Cursor Settings" and press Enter
Navigate to Models

Step 2: Select GPT-5 Codex

In the Models section, you will see a list of available models. Enable gpt-5-codex if it is not already enabled:

Available Models:
  [x] claude-sonnet-4              (Anthropic)
  [x] claude-opus-4                (Anthropic)
  [x] gpt-5-codex                  (OpenAI)    <-- Enable this
  [x] gpt-4o                       (OpenAI)
  [x] gemini-2.5-pro               (Google)
  [ ] deepseek-v3                   (DeepSeek)

Step 3: Use in Chat or Composer

Open the Cursor chat panel (Cmd+L) or Composer (Cmd+I) and select gpt-5-codex from the model dropdown at the top of the panel.

Model: gpt-5-codex ▼
─────────────────────
You: Refactor the authentication module to use refresh tokens

Method 2: Use Your Own OpenAI API Key

If you want to bypass Cursor's request limits or use GPT-5 Codex on the free plan, you can bring your own API key.

Step 1: Get an OpenAI API Key

Go to platform.openai.com
Navigate to API Keys
Click Create new secret key
Copy the key (starts with sk-)

Step 2: Configure in Cursor

Open Cursor Settings (Cmd+Shift+P > "Cursor Settings")
Navigate to Models > OpenAI
Paste your API key in the OpenAI API Key field

OpenAI Configuration:
  API Key: sk-xxxxxxxxxxxxxxxxxxxxxxxx
  Base URL: https://api.openai.com/v1  (default)

Step 3: Add GPT-5 Codex as a Custom Model

If gpt-5-codex does not appear in the default list, add it manually:

In Models settings, click + Add Model
Enter the model ID: gpt-5-codex
Set the provider to OpenAI
Save

You can now select it from any model dropdown in Cursor.

Method 3: Use via OpenRouter (More Model Options)

OpenRouter acts as a proxy that gives you access to multiple providers through a single API key. This is useful if you want to switch between GPT-5 Codex and other models seamlessly.

Step 1: Get an OpenRouter API Key

Go to openrouter.ai
Create an account and add credits
Copy your API key

Step 2: Configure OpenRouter in Cursor

Open Cursor Settings > Models
Under OpenAI API Key, enter your OpenRouter key
Change the Base URL to:

https://openrouter.ai/api/v1

Step 3: Add the Model

Add the OpenRouter model ID:

openai/gpt-5-codex

Now you can access GPT-5 Codex through OpenRouter's proxy, along with any other model they support.

Practical Usage Examples

Example 1: Generate a REST API

Open Composer (Cmd+I) with gpt-5-codex selected:

Prompt: Create a complete REST API for a task management app using Express and TypeScript.
Include:
- CRUD endpoints for tasks
- Input validation with Zod
- Error handling middleware
- SQLite database with Drizzle ORM
- Unit tests with Vitest

GPT-5 Codex will generate a multi-file implementation that you can review and apply.

Example 2: Debug a Complex Issue

In Cursor chat (Cmd+L):

Prompt: I'm getting a race condition in my WebSocket handler.
Multiple clients can update the same resource simultaneously
and the last write wins, causing data loss.

@src/ws/handler.ts
@src/services/resource.ts

Help me implement optimistic concurrency control.

Example 3: Refactor Legacy Code

Prompt: Refactor this jQuery-based module to modern React with TypeScript.
Preserve all existing functionality and add proper types.

@src/legacy/dashboard.js

Example 4: Generate Tests from Implementation

Prompt: Write comprehensive tests for this module.
Cover happy paths, error cases, boundary conditions, and concurrent access.
Use Vitest. Aim for > 90% coverage.

@src/services/payment.ts

Optimizing GPT-5 Codex Performance in Cursor

Use .cursorrules Files

Create a .cursorrules file in your project root to give GPT-5 Codex project-specific context:

You are working on a Next.js 15 application.

Tech stack:
- TypeScript strict mode
- Tailwind CSS v4
- Prisma ORM with PostgreSQL
- NextAuth.js for authentication
- Vitest for testing

Conventions:
- Use server components by default
- Use "use client" only when necessary
- Prefer named exports
- Use Zod for all input validation
- Error boundaries for every route segment

When generating code:
- Always include proper TypeScript types
- Add JSDoc comments for public functions
- Include error handling
- Write tests alongside implementation

Optimize Context Usage

GPT-5 Codex has a 200K token context window, but using it efficiently still matters:

Technique	How
Reference specific files	Use `@file` to include only relevant files
Use folders selectively	`@src/services` instead of `@src`
Exclude large files	Avoid referencing generated files, lock files, or bundled code
Break tasks down	Multiple focused prompts beat one massive prompt
Use codebase indexing	Let Cursor's indexer provide automatic context

Model Switching Strategy

Different models have different strengths. Use Cursor's model dropdown to switch based on the task:

Task	Recommended Model	Why
Complex architecture	Claude Opus 4	Best reasoning
Code generation	GPT-5 Codex	Optimized for code
Quick edits	GPT-4o	Fast and cheap
Long-context tasks	Gemini 2.5 Pro	1M token window
Debugging	Claude Sonnet 4	Strong analysis

GPT-5 Codex vs Other Models in Cursor

Here is how GPT-5 Codex compares to other popular models for coding tasks in Cursor:

Benchmark	GPT-5 Codex	Claude Sonnet 4	Gemini 2.5 Pro
SWE-Bench (code fix)	72%	70%	65%
HumanEval (code gen)	96%	94%	92%
Multi-file edit accuracy	Excellent	Excellent	Good
Speed (tokens/sec)	~90	~80	~100
Cost efficiency	Medium	Medium	High

Benchmarks are approximate and based on publicly available evaluations.

Troubleshooting

Issue	Solution
Model not appearing	Add it manually in Settings > Models > Add Model
"Rate limit exceeded"	Wait a few minutes, or switch to your own API key
Slow responses	Check network; try a different model temporarily
Poor code quality	Add a `.cursorrules` file with project context
"Invalid API key"	Verify key is correct and has sufficient credits
Context too long	Use `@file` references instead of pasting code

Conclusion

GPT-5 Codex paired with Cursor AI creates one of the most productive coding setups available in 2026. Whether you use Cursor's built-in allocation, your own OpenAI key, or OpenRouter, the setup is straightforward and the results are impressive for code generation, refactoring, and debugging tasks.

If your project involves AI-generated media such as images, video, or audio, Hypereal AI provides affordable, developer-friendly APIs that you can integrate directly into your Cursor-powered workflow. Build and test your media generation pipelines faster with production-ready endpoints.