Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- GPT-5-Codex introduces agentic coding with cloud hand-offs.
- GitHub integration catches bugs and backward compatibility issues.
- Usage of Codex surged 10x among developers in a month.
OpenAI today announced that GPT-5, its latest-generation large language model, has been optimized for coding and is now available in Codex.
(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
This is part of an ongoing series of improvements and upgrades that OpenAI has introduced recently for Codex. Here is a short list:
“Codex performed best in our backend Python code-review benchmark. It was the only one to catch tricky backward compatibility issues and consistently found the hard bugs that other bots missed,” says Aaron Wang, senior software engineer at Duolingo.
Trained on actual coding tasks
OpenAI said that it was trained on actual coding tasks, including building full projects from scratch, adding features, incorporating test validations, actual debugging, doing large-scale refactors (restructuring, consolidating, moving, renaming, but in a way that nothing breaks), and code reviews. It also follows prepared instructions presented in the AGENTS.md file. AGENTS.md is the name of a file that is kind of a readme for AIs.
Also: How to use ChatGPT to write code – and my top trick for debugging what it generates
It’s trained on complex, real-world engineering tasks such as building full projects from scratch, adding features and tests, debugging, performing large-scale refactors, and conducting code reviews. According to OpenAI, it’s more steerable, adheres better to AGENTS.md instructions, and produces higher-quality code — just tell it what you need without writing long instructions on style or code cleanliness.
“I needed to update a codebase owned by another team for a feature release. Landing the change would have usually meant making a choice between taking the time to refactor and add tests and missing the deadline, or doing things quickly and risking instability,” reported Tres Wong-Godfrey, tech lead at Cisco Meraki.
“With Codex,” he said, “I offloaded the refactoring and test generation while focusing on other priorities. It produced high-quality, fully tested code that I could quickly hand back — keeping the feature on schedule without adding risk.”
Nothing short of extraordinary
Codex’s new integration with the VS Code development environment has become my personal favorite drug of choice in the last week. That addition has supercharged my coding considerably. I mean, considerably. I described my first weekend using Codex in VS Code. I got roughly 24 days of coding done in 12 hours.
Also: I did 24 days of coding in 12 hours with a $20 AI tool – but there’s one big pitfall
Since then, I upgraded from the $20 Plus tool to the $200-per-month Pro tool. I’m still doing a lot of mental and emotional processing about what I got done in the last four days. I’ll have a full article on that soon. Suffice it to say that the pace of work output has been nothing short of extraordinary, and the dopamine rush I get at completing so much work so fast is surprisingly addictive.
Interesting insights
The model that OpenAI has announced is called GPT-5-Codex. The way OpenAI explains it: “Unlike GPT-5, which is a general-purpose model, we recommend using GPT-5-Codex only for agentic coding tasks in Codex or Codex-like environments.” The company has an anecdotal report from someone who gave it a programming challenge, and the LLM ran autonomously for more than seven hours.
Also: The best AI for coding in 2025 (including a new winner – and what not to use)
I talked with OpenAI’s Codex project lead on Friday and got some interesting insights about where Codex might fit into your programming stack. AI coding tools tend to come in a variety of general forms:
- Command-line interfaces: You control the AI experience all in a terminal window. For power CLI programmers, this is the way to go.
- Agentic coding partners: You give the agents assignments and they go off and do their thing, reporting back when they’re done. These often live in GitHub.
- Chatbot assistants in the IDE: These are often a combination of some of the other functions, like performing some of the command-line tasks but in their own window, and doing agentic tasks as well.
- Code completion assistants: These live right in the editor and help complete coding as you type, sometimes answering or writing code with you.
- No-code AI tools: Products that build applications out of building blocks, often with an AI assist. Think web page builder with some smarts, but also for apps.
Let’s ignore the no-code tools, because they’re often proprietary, limited, and don’t actually write code.
Codex is tuned to do the first two: provide command-line support and be an agentic coding partner. It added IDE support, but only of the chatbot variety. Codex doesn’t participate in the code-completion area of IDE expansion.
Also: ChatGPT just saved me 25% off my dinner tonight – here’s how
When I asked the Codex product lead about it, he said they were working very hard to make sure Codex would work with code completion tools in the same IDE. So if you want to use GitHub Copilot and Codex, or Cursor and Codex, you’ll be good to go.
One thing I really like about OpenAI’s recent introductions is that all of this runs through your ChatGPT account. Many of the other tools use API keys and calls. In fact, GPT-5 in some of the other tools is driven by API calls. So OpenAI’s new offerings will be more convenient, saving you the hassle of setting up, configuring, and linking APIs to applications (sometimes over and over again).
You need a separate OpenAI account for those API calls, and you’re billed based on the number of tokens you use. I use my OpenAI API account for a project when I’m making calls into the API and getting back programmatic results. I use a couple of hundred tokens every time.
Also: I tested GPT-5’s coding skills, and it was so bad that I’m sticking with GPT-4o (for now)
But by integrating directly into the ChatGPT account, you get access to all of the ChatGPT features, along with the Codex features, all with a single login experience. It also means that if you upgrade to Pro (as I did recently), you not only get an expanded Codex usage allotment, you also get the Pro ChatGPT and Deep Research features as well.
So much more predictable
Like I said, when you use API calls, you’re billed on usage. But if you subscribe to either the Plus or Pro account, you know how much you’re going to spend per month.
OpenAI says they’re seeing strong adoption of Codex among developers and saw usage grow by 10 times in just the past month. I can believe it. This is starting to get really real.
Also: How I used ChatGPT to analyze, debug, and rewrite a broken plugin from scratch – in an hour
Stay tuned. I’ll tell you more about my recent experiences soon.
What do you think about OpenAI’s new GPT-5 Codex and its focus on agentic coding? Have you tried using AI tools directly inside your IDE, or do you prefer code completion assistants like GitHub Copilot? Do you see yourself handing off longer coding sessions to the cloud, or does that feel risky? And with usage reportedly growing so fast, do you expect tools like this to become standard in most developers’ workflows soon? 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.
Leave a Reply