Video tutorial · The Metaverse Guy
Codex vs Claude Sonnet 4.5: The Best AI Coding Agent in 2025?
I put OpenAI Codex CLI through its paces against Claude Sonnet 4.5, Cursor, and other tools to see which AI coding assistant really delivers.
- Published
- 29 September 2025
- Length
- 9 min
- Views
- 2,504 views
In this video, I test the newly re-released OpenAI Codex CLI and compare it directly with Claude Sonnet 4.5, GitHub Copilot, Google Jules, and Cursor. I walk through installing Codex CLI, generating a dashboard project in the terminal, using it inside VS Code, and trying the browser-based GitHub integration. The goal is to give you a practical, no-hype verdict on whether Codex can replace your current coding assistant.
What is OpenAI Codex and who can use it?
OpenAI Codex is a command-line and IDE-based coding agent that has recently been updated and made available again. It is not a new model but a tool that orchestrates code generation, debugging, and automation using OpenAI’s GPT-5 models. You can access it through the terminal, inside editors like VS Code, Cursor, and Windsurf, or via a cloud-based browser interface.
At the time of recording, Codex requires a paid subscription. You can find the official page and availability details on OpenAI’s site. It is not yet open to all free-tier users.
Installing Codex CLI step by step
I started by installing Codex CLI on my machine. The process is straightforward but does require a login and an active subscription. Here is exactly what I did.
- Open your terminal and run the install command provided on the official Codex page.
- Once installed, launch Codex and complete the first-time login. You will be prompted to authenticate with your OpenAI account.
- Verify that your account has an active subscription, as Codex will not work without one.
- Create a new project folder using the Codex CLI init command to set up a workspace.
Building a dashboard project with Codex CLI
With the CLI ready, I asked Codex to generate a dashboard project directly in the terminal. The tool started scaffolding HTML, CSS, and JavaScript files, and I had to approve each step as it went along. The execution flow pauses frequently for user confirmation, which gives you control but slows things down.
The entire process took over 10 minutes to complete. While the final output was functional, the speed felt sluggish compared to what I am used to with IDE-based agents. If you are working on larger projects, this approval-heavy flow might become a bottleneck.
Using Codex inside VS Code
I then switched to the VS Code extension to see if the experience improved. After installing the Codex extension, I opened the chat interface and asked it to generate the same dashboard. This time, the generation was noticeably faster and the chat made it easier to iterate on the code.
Comparing the CLI and IDE results, the VS Code version produced a cleaner output in less time. The chat interface also lets you refine the code without leaving the editor, which feels more natural for day-to-day development.
Codex in the browser with GitHub integration
Codex also works directly in the browser, where you can connect it to a GitHub repository. I tested this by running a code review on an existing project. The browser interface pulled in the repo, analysed the code, and provided suggestions inline. This workflow is handy for quick reviews or when you are not at your own machine, but it is not a full replacement for a local development environment.
Codex vs Claude Sonnet 4.5 and other tools
Throughout these tests, I kept comparing Codex against Claude Sonnet 4.5, GitHub Copilot, Google Jules, Gemini, and Cursor. Claude Sonnet 4.5 remains a very strong coding assistant, especially for understanding complex codebases and generating accurate snippets. Cursor, with its tight editor integration, still feels more responsive and polished for everyday coding tasks.
Codex’s main limitation right now is that it only works with GPT-5 models. You cannot swap in Claude or other models, which restricts flexibility. Tools like Cursor and the Claude Code CLI let you choose your model, which is a significant advantage when different tasks benefit from different AI backends.
My verdict: is Codex the best AI coding agent?
After testing Codex CLI, the VS Code extension, and the browser version, I cannot say it beats Claude Sonnet 4.5 or Cursor outright. It is a capable tool, but the CLI is slower than I would like, and the model lock-in is a drawback. For terminal purists who want an OpenAI-native workflow, it might fit, but most developers will find Cursor or Claude Code more productive right now.
Codex is worth watching as it matures, but for the moment, I am sticking with my current setup. If you are curious, give it a try—just be prepared for a slightly rough experience compared to the more established alternatives.
Tools used in this video
- OpenAI Codex CLI
- Claude Sonnet 4.5
- GitHub Copilot
- Google Jules
- Cursor AI
- Claude Code
- Gemini
- Google AI Studio
- VS Code
- Windsurf
- GPT-5
Questions
- How do I install Codex CLI?
- Run the install command from OpenAI’s official Codex page in your terminal, then log in with your OpenAI account. You need an active paid subscription to use it.
- Is Codex CLI free?
- No, at the time of recording Codex requires a paid subscription. It is not available on the free tier.
- Can I use Codex with Cursor or VS Code?
- Yes, Codex offers extensions for VS Code, Cursor, and Windsurf. The IDE experience is generally faster and more interactive than the CLI.
- How does Codex compare to Claude Sonnet 4.5?
- In my tests, Claude Sonnet 4.5 still holds its own for code understanding and generation. Codex is promising but limited to GPT-5 models and feels slower in the terminal. Cursor and Claude Code currently offer a more polished experience.
Chapters
- 00:00 Intro: OpenAI Codex vs Claude Code
- 00:13 What Codex is & who can access it
- 00:32 Codex official page & availability
- 00:45 Using Codex in IDEs (VS Code, Cursor, Windsurf)
- 00:55 Codex in terminal & cloud version overview
- 01:29 Installing Codex CLI step by step
- 02:04 First login & subscription requirement
- 02:24 Creating a new project folder with Codex CLI
- 02:41 Generating a dashboard project in terminal
- 03:44 Codex CLI execution flow & approval commands
- 04:43 Codex creating dashboard HTML, CSS, and JS
- 05:35 Reviewing CLI performance & time taken (~10+ mins)
- 05:54 Switching to IDE usage (VS Code demo)
- 06:33 Codex chat interface walkthrough
- 07:05 Dashboard generation inside VS Code
- 07:45 Comparing Codex CLI vs Codex IDE results
- 08:08 Codex limitations (only GPT-5 models)
- 08:36 Using Codex inside the browser with GitHub integration
- 09:20 Code review example using Codex in browser
- 09:53 My personal verdict on Codex vs Cursor
- 10:20 Conclusion & recommendations
Video description and links
This article was written from the video's description with the help of AI. Watch the video for the full walkthrough.



