# Claude Code vs. Codex: Why I use both and how to choose one (if you must) > Source: > Published: 2026-08-12 13:49:33+00:00 # Claude Code vs. Codex: Why I use both and how to choose one (if you must) *Follow ZDNET: *[Add us as a preferred source](https://cc.zdnet.com/v1/otc/00hQi47eqnEWQ6T9d4QLBUc?element=BODY&element_label=Add+us+as+a+preferred+Google+source&module=LINK&object_type=text-link&object_uuid=5e5d2e64-4b30-43e6-8555-26eac7e449f3&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=379e95d2-6b56-476b-a90b-043a8dd63bd3)* on Google.* ### ZDNET's key takeaways - Claude Code and Codex are more alike than different. - Your workflow and ecosystem matter more than benchmarks. - Trying both may be smarter than choosing one. Given their incredible level of adoption, it's hard to believe that agentic coding systems like Claude Code and OpenAI Codex have only been with us for about 18 months. In that time, they've gone from promising novelties to amazing and highly productive members of our programming teams. For those of us who have already experienced either or both of these tools, they've become mission-critical elements of our overall programming stack. But for those of you who haven't yet experienced these tools, which one should you choose? What's the difference? Why choose one over the other? **Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned** These are the questions this article aims to answer. I've used both extensively, so I'll share some of my personal experiences. ## They're more similar than different Both tools do essentially the same job. They build software, often from start to finish. Early on, AI coding tools were characterized by smart autocomplete, where the AI would complete lines or blocks of code for you. But now, it's different. I just talk to Claude and Codex and they build entire systems. For me, this is almost more natural than writing code. As much as I've been coding since the days of punch cards, I've also managed programmers as both a product manager and an engineering manager. So specifying a project and managing a team is second nature to me. That's what working with Claude Code and Codex is like. Don't think line-by-line. Think assignment by assignment. It's almost misleading to call these tools. Let's use workshop tools as an analogy. For decades, we were at the hand tool level. We used pull saws to cut wood. Then, when generative AI hit the scene, we jumped to the power tool level. Instead of using muscle power to cut wood, we used power saws. **Also: Worried about AI coding? Why the invention of power tools is the blueprint for your career future** Now we're at the general contractor level. Instead of using any tools at all, we just tell our contractors what to build, and they do. Here's one more mindset change for you. For the first few years after generative AI hit, I ran articles comparing various AIs, and rated which ones could generate working code. That is no longer an issue. Both of the AIs I discuss in this article are very good at producing quality code. If something doesn't work, it's because they got confused or didn't receive good direction. In many ways, it's not even a cost issue. Although there's some nuance to how each AI charges for use, most individual developers will pay just about the same for each tool. At this point, which tool you choose has more to do with which ecosystem you're more familiar with, which company you prefer to work with, and which environment works best for you. Some of you might even choose to use both AIs. ## So what is agentic coding? Put very simply, chatbots answer. Agents do stuff. So when you ask ChatGPT to write a block of code, it will give you back a block of code in a text box. Codex will actually build all the parts of the software for you, test it, and make it work. Both Claude Code and Codex can work on existing projects, modify files, run commands on the computer, run tests, find and fix bugs, review their work, evaluate output, and run for a long time without needing much input. Both tools can run multiple projects at once, using individual agents to work on either different pieces of one problem or multiple problems at the same time. Both now also work with other tools and resources through connectors and plugins. ## How I use them I have a couple of terminal windows open while I'm writing this. Claude Code is currently attempting to emulate a variety of drawing styles for a project I'm working on. It's been coding away, making changes, looking at its work, comparing its output to samples of human drawings, and occasionally sharing its output with me to get my input. **Also: How I set up Claude Code in iTerm2 to launch all my AI coding projects in one click** I keep the terminal window open on a side monitor, so I can actually see Claude at work while I'm working on a completely different project. This very much mirrors the management style I used when I had direct reports. Throughout my day, I'd be working on my projects. But I'd regularly help out and review my team's work. Back in September 2025, I started with Codex. I used it to do some massive work with my commercial WordPress security plugins. I spent $200 for a month, and got an incredible amount accomplished. But once that code shipped, I dropped back to the $20/mo ChatGPT plan, which is where I've been ever since. I then moved to Claude. I didn't do that because Claude was better, but because I wanted experience with it to report back to you. Rather than work on the same code with both, I started new projects with Claude. In Claude's case, I'm using it to create two different Apple ecosystem apps, both of which run on iPhone, iPad, Mac, and the Apple Watch. I've been using the $100/month Max plan with Claude, which has given me enough AI cycles to get my work done. I've also found myself throwing little side projects at Claude instead of Codex, mostly because I didn't want to spend another $100/month on the higher Codex tier. To be fair to ChatGPT, you do get a couple of good five-hour coding sessions in Codex for just your $20. I [wrote about that a month or so ago](https://www.zdnet.com/article/massive-wordpress-spam-attack-stopped-with-codex-and-claude/). I used Codex to do a major security mitigation, all on that twenty bucks. Although both Claude Code and Codex have desktop apps, I don't use them much for coding work. I use Claude Code almost entirely in the terminal because I can color-code my terminal sessions and it has tabs. I tried using both [Codex and Claude Code inside Apple's Xcode](https://www.zdnet.com/article/how-to-create-your-first-iphone-app-with-ai-no-coding-experience-needed/) and it was unpleasant. But the terminal interface rocks. On the other hand, I use Codex almost exclusively [inside VS Code](https://www.zdnet.com/article/why-im-switching-to-vs-code-hint-its-all-about-ai-tool-integration/). It fits with my project and works well for testing and debugging. (A quick note to Anthropic and OpenAI: If you want more people to use Claude Code or Codex in your desktop apps, give us tabs and let us customize the colors, fonts, and text sizes like we do in terminal.) If there is one thing you can take away from this article about my experience, it's that I don't have a favorite. Both Claude Code and Codex work exceptionally well. I use one or the other depending on the project I'm working on, not because one does more or is better than the other. Speaking of, for that major security mitigation, I used Codex and Claude together. As it turns out, using the two AIs simultaneously to work on different parts of the problem is brilliant. Incredibly effective. So don't just think about which to choose. You might want to use both. ## Understanding the ecosystems In the beginning, there was chat. And it was good-ish. But that was the product, whether in the form of ChatGPT or Claude, that burst onto the scene in late 2022. Later, in 2025, agentic coding tools like Claude Code and Codex began to take form. In the early days, the agentic coding tools had completely separate user interfaces from the chatbots. Over time, some intrepid users started using the agentic coding tools to do non-coding work. After all, they could manipulate files, do long sets of steps, and plan their work. The AI companies noticed, and earlier this year, they productized this non-coding off-label use of the coding agents. First came [Claude Cowork](https://www.zdnet.com/article/anthropic-claude-cowork-ai-file-management-test/) and then [ChatGPT Work](https://www.zdnet.com/article/chatgpt-work-claude-cowork/). **Also: I let ChatGPT Work and Claude Cowork loose on my files - only one made me nervous** For a short time, these were also separate tools. But just a few months ago, both OpenAI and Anthropic combined their chat, agentic coding, and agentic work tools into single desktop applications. Personally, I think that this almost-forced bundling [nerfed some of the value](https://www.zdnet.com/article/openai-gutted-chatgpt-desktop-app-for-codex-work/) of the individual apps. But that's progress, right? That means if you want to use either agentic coding tool, you're really buying a subscription to the whole ecosystem. To be honest, it's not a bad deal. I mentioned how much work I got done with just the $20/month ChatGPT Plus subscription, which comes with ChatGPT, Codex, and ChatGPT Work. My $100/month Claude Max subscription includes the Claude chatbot, Claude Code, and Claude Cowork. There's a lot of value to this, because you can use your usage allocation across project types. Of course, the more you use, the more likely you're going to run into usage walls, and therefore the more likely that you'll need to upgrade. And that's what the AI companies are hoping for -- cha-ching! **Also: I let Anthropic's Claude Cowork loose on my files, and it was both brilliant and scary** In any case, this is why ecosystem matters when picking Claude Code or Codex. If you're already paying for ChatGPT, you'll probably just use Codex. If you're already paying for Claude, Claude Code is a no-brainer. Both come along for the ride with the subscription. ## But are there any differences? Both have worktrees, cloud execution, scheduled routines, a desktop GUI, browser automation, mobile remote control, skills, subagents, skill record-and-replay, conversational mode, parallel tasks, and even some version of self-hosted sites generated by the AI. To me, there is one standout feature that distinguishes ChatGPT from Claude: image generation. ChatGPT has ChatGPT Images, which is very impressive. I actually fed into it my Mac app's user interface from my Claude Code project and ChatGPT gave me back [a substantially improved user interface mockup](https://www.zdnet.com/article/i-asked-chatgpt-to-redesign-my-apps-wow-results/). Even if you're a Claude Max subscriber, it's worth buying a $20/month ChatGPT Plus plan just for this one feature alone. It's a great complement to Claude Code. **Also: I asked ChatGPT Images 2.0 to redesign my app UIs - and wow** Although I haven't tried this, Claude can run very large multi-agent jobs. Since [Opus 4.8](https://www.zdnet.com/article/anthropics-launches-opus-4-8-where-honesty-is-the-killer-feature/), it has been able to use Dynamic Workflows to split huge tasks among hundreds of AI agents and then coordinate the results. For enterprises, Claude can help you preserve privacy by [running Claude in self-hosted cloud environments within your own corporate infrastructure](https://code.claude.com/docs/en/self-hosted-environments). As far as I know, Codex doesn't offer a self-hosted cloud environment yet. From a small coding team's perspective, both environments are enormously powerful. Once you start looking at enterprise deployment, definitely talk in depth with Anthropic and OpenAI to get a feel for the unique capabilities they can offer to your enterprise. ## What's it all mean? Anthropic and OpenAI have been aggressively building up both of their offerings, but because the AI space moves at such a rapid pace, it's difficult to nail down what features exist now vs. which ones have changed within the last few weeks. Likewise, just as soon as one company makes a move, its competitor makes a similar move. Giving you a clear presentation about why one is better than the other, or which one you should choose, is almost impossible because they are both incredibly capable tools. **Also: Anthropic launches Opus 4.8, with honesty as its killer feature** If you prefer using ChatGPT as a chatbot over Claude as a chatbot, then use Codex. If your focus is primarily on programming, pick the environment you prefer. And while the desktop apps are a major focus for the AI companies themselves, it's really the integration into your work style that makes the most sense. Look at the apps, but frankly, [they're not as good as they could be](https://www.zdnet.com/article/openai-gutted-chatgpt-desktop-app-for-codex-work/). Instead, choose based on whether you prefer them in terminal or in VS Code. There are also browser, cloud, and mobile interfaces. Codex is available (although with heavy limits) in the ChatGPT free tier. For Claude Code, you'll need to buy the $20/month tier to use it at all. I recommend you start with both AIs on the lower tiers, get some experience with each, and see which one you like better. If you have the budget, I would also suggest picking one and using it on a higher-cost tier, and then pick the other AI and use it with a less expensive plan. That will give you the best of both worlds. If you have to pick one, and you're moving above the $20/month chatbot expense, I would recommend going with the one that includes all three features you prefer. Personally, I enjoy working with both. My choice (if I had to choose) would depend entirely on which one the project I'm working on had more history in. If I were continuing with Apple programming, then I would choose Claude because that's where all my Apple programming history is embedded in the project's files. If I were focusing on additional WordPress plugins, I would go with Codex because that's the tool I used to develop them. If you're starting from scratch, start with the chatbot you like most, and then add a little agentic coding. See if you like how it works, and then try the other. Whichever one you gravitate to will likely be your choice. I know that's not as concrete a recommendation as I would normally give, but these tools are locked into such a neck-and-neck battle for dominance that, in some ways, they're almost indistinguishable. The good news is that they're both fairly affordable and astonishingly good. That's the world of AI. Welcome to the future. Which AI have you decided deserves your money: Claude Code or Codex? Let us know in the comments below. *You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.* #### Featured [Editorial standards](/editorial-guidelines/)