AI-Powered Coding with OpenAI Codex: Features, Install & VSCode Guide

Codex Overview

Codex is an OpenAI‑released software engineering intelligence agent built on the codex-1 model. It uses natural language to generate code, fix bugs, and submit pull requests, running tasks in an isolated cloud sandbox and supporting multiple entry points such as CLI, IDE, and GitHub.

Key Features

Product Form & Integration : Cloud‑agent architecture with parallel task execution and pre‑installed code libraries to avoid local risks.

Automation : Automatic PR generation and submission, high merge rate (>80%).

Multi‑Entry Collaboration : Supports ChatGPT, CLI, IDE, and GitHub with shared context.

Programming Efficiency : Optimized codex-1/GPT‑5‑Codex model, SWE‑bench verified 74.5% accuracy, supports 12+ languages, reduces token consumption by 93.7% compared to GPT‑5.

Safety & Reliability : No external network access by default, multi‑layer prompt injection defenses, PRs require manual confirmation.

Workflow & UX : IDE can read selected code as context, minimal prompt construction overhead, no container restarts needed for small changes.

Feature Comparison with Claude Code

Codex generally outperforms Claude Code in budget efficiency, task completion, code quality, deep‑thinking ability, large‑scale refactoring, multi‑task handling, and long‑context support, while Claude Code has advantages in tool‑specific safety checks and certain UI rendering.

Installation

Install Codex CLI globally via npm: npm install -g @openai/codex Configure the CLI by creating $HOME/.codex/config.toml (or C:\Users\<username>\.codex\config.toml on Windows) with the desired model provider and API keys. Example OpenRouter configuration:

# Select LLM Provider
profile = "openrouter"

[model_providers.openrouter]
name = "OpenRouter"
base_url = "https://openrouter.ai/api/v1"
env_key = "OPENROUTER_API_KEY"
wire_api = "chat"

[profiles.openrouter]
model = "openai/gpt-5-codex"
model_provider = "openrouter"
model_reasoning_effort = "high"

Set the environment variable for the API key (Linux/macOS): export OPENROUTER_API_KEY="sk-or-v1-..." On Windows, add the variable via system settings or use:

set OPENROUTER_API_KEY="sk-or-v1-..."

Basic Usage

After configuration, run the CLI: codex On Windows, WSL may be required for shell commands.

Common Commands

/init

: Generate or update AGENTS.md with repository conventions. /status: Show current model and approval mode. /approvals: Switch between read‑only and automatic approval. /model: Change the underlying LLM and reasoning effort. /review: Perform a quick code review based on git diff. /new: Start a new conversation without losing existing context. /diff: Display git differences, including untracked files. /mention: Reference a file in the conversation without pasting its content. /compact: Compress the dialogue to stay within token limits. /logout: Log out of Codex. /quit: Exit the CLI.

VSCode Integration

Install the Codex extension from the VSCode marketplace, then authenticate via the ChatGPT login (ChatGPT Plus membership works). The extension shares the same ~/.codex/config.toml configuration as the CLI, allowing seamless transition between terminal and editor.

After login, you can interact with Codex directly within VSCode, similar to other AI‑assisted development extensions.

Model Options

OpenRouter provides a wide range of models, including Anthropic Claude series, OpenAI GPT‑5 variants, Google Gemini, X‑AI Grok, Z‑AI GLM‑4.6, Moonshot Kimi, Qwen, and DeepSeek. You can also configure ModelScope providers for domestic models such as Qwen‑3‑Coder‑480B‑A35B.