OpenAI introduces GPT-5.3-Codex and raises stakes for coding AI — OpenAI / Elyse Betters Picaro / ZDNET

Follow ZDNET: Add us as a preferred source on Google.

ZDNET's key takeaways

GPT-5.3-Codex helped debug and deploy parts of itself.
Codex can be steered mid-task without losing context.
"Underspecified" prompts now produce richer, more usable results.

OpenAI today announced the launch of GPT-5.3-Codex, which OpenAI claims is the most capable agentic coding model yet released. Interestingly enough, Anthropic, maker of Claude Code, is also releasing a new powerful model at exactly the same time. Anthropic's model, Opus 4.6, is a more general-purpose model, but Anthropic is well known for its Claude Code offering.

Also: OpenAI's Codex just got its own Mac app - and anyone can try it for free now

As with Anthropic's Opus 4.6, OpenAI describes GPT-5.3 as a "frontier model." This is a term the AI industry uses to describe models at the bleeding edge of performance and capability. According to OpenAI, the new 5.3 Codex runs 25% faster, enabling longer-running tasks.

OpenAI's GPT-5.3-Codex is being announced just days after OpenAI's announcement and release of a Mac app dedicated to Codex.

Astute watchers of OpenAI's GPT models will notice the 5.3 version number on this release. OpenAI's current hottest GPT release is GPT-5.2, not GPT-5.3. Although nothing has been announced or even hinted at, I'm guessing we're not too far away from a general GPT-5.3 release in the next few days or weeks.

Coding AI, code thyself

I'm not sure if this is incredibly exciting or deeply disturbing. GPT-5.3-Codex is OpenAI's "first model that was instrumental in creating itself." The Codex team used Codex to debug its own training, manage its own deployment, and diagnose test results.

Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned

OpenAI's blog post says, "With GPT-5.3-Codex, Codex goes from an agent that can write and review code to an agent that can do nearly anything developers and professionals can do on a computer." Let's just hope it doesn't decide that humans are superfluous.

More than just coding

Acknowledging that developers, designers, product managers, and data scientists do a lot more in their jobs than just code, GPT-5.3-Codex is designed to support "all of the work in the software lifecycle -- debugging, deploying, monitoring, writing PRDs, editing copy, user research, tests, metrics, and more." The company reports that it can help you build whatever you want to build, specifically mentioning building slide decks and spreadsheets.

Also: Want local vibe coding? This AI stack replaces Claude Code and Codex - and it's free

The company says that this model is designed to be steered mid-task, with continuous interaction and no loss of context. The new Codex Mac app (and, presumably, an upcoming Windows app) can definitely help in maintaining interaction with the model as it works on projects for you.

The span of operation is also much longer. Codex can run processes that take more than one day. As a test, OpenAI built two web-based games. Using the "skill" feature introduced with the Mac app, testers used a web gaming development skill to build the two games over millions of tokens.

The blog post specifically highlighted how GPT-5.3-Codex better understands intent. If you're making a "day-to-day website," the blog post reports that "Simple or underspecified prompts now default to sites with more functionality and sensible defaults, giving you a stronger starting canvas to bring your ideas to life."

In other words, if you ask Codex to "make me something pretty," it will. I had some significant success asking Codex 5.2 to redesign some parts of my WordPress security plugin to make the administration screens more attractive and compelling. Some of the design features that existed in earlier Codex versions have been improved upon in this latest release.

Other examples cited by the blog include dynamic pricing displays and automated testimonial carousels.

Frontier benchmarks and technical performance

One of the subtexts I picked up in the technical briefing I participated in last week for the Mac Codex app was the desire on the part of some of its customers for faster response times. I know that I often set Codex or Claude Code off on a programming task and then switch to some other activity while waiting for the process to complete.

Also: 10 ChatGPT Codex secrets I only learned after 60 hours of pair programming with it

While my experience has been that agentic coding speeds up my development process from months to days, that wait is still a bit painful. So increasing the performance, and thereby the output speed of the models, is always a win.

According to the blog post, "GPT-5.3-Codex sets a new industry high on SWE-Bench Pro and Terminal Bench." The company says assignments take fewer tokens, increasing efficiency. OpenAI also says, "GPT-5.3-Codex sets a new industry high on SWE-Bench Pro and Terminal Bench, and shows strong performance on OSWorld and GDPVal." These are benchmarks the company uses to measure coding, agentic operations, and real-world capabilities.

Security, infrastructure, and safeguards

OpenAI has a published Preparedness Framework, which it uses for measuring and protecting against severe harm from frontier AI capabilities. In the blog post announcing GPT-5.3-Codex, the company says that this new model is the first one it classifies as "high capability" for cybersecurity tasks.

It's been trained to identify software vulnerabilities, with expanded safeguards and monitoring. Concurrently, the company is launching Trusted Access for Cyber, a pilot program to accelerate cyber defense research, and is donating $10M in API credit grants to support cybersecurity research. If your organization is engaged in "good faith security research," you can apply for API credits through the company's Cybersecurity Grant Program.

Also: Is ChatGPT Plus still worth your $20? I compared it to the Free, Go, and Pro plans - here's my advice

Supporting these actions, the blog post reports, "We're taking a precautionary approach and deploying our most comprehensive cybersecurity safety stack to date. Our mitigations include dual-use safety training, automated monitoring, trusted access for advanced capabilities, and enforcement pipelines, including threat intelligence."

Availability

GPT-5.3-Codex is available now with paid ChatGPT plans across the Codex app, CLI, IDE extension, and web. API access is planned, and the company expects to deliver 25% faster interactions. The Codex usage promotion we reported earlier this week for free ChatGPT users still stands, but they'll be limited to GPT-5.2-Codex for now.

Also: Stop using ChatGPT for everything: My go-to AI models for research, coding, and more (and which I avoid)

What do you think about GPT-5.3-Codex and the direction OpenAI is taking with more autonomous, long-running coding agents? Have you tried Codex or similar tools like Claude Code in real development work? If so, did the speed or autonomy change how you work? How do you feel about an AI model helping debug, deploy, and even participate in its own development? Do the cybersecurity safeguards and "frontier model" positioning reassure you, or raise new concerns? 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.

OpenAI's new GPT-5.3-Codex is 25% faster and goes way beyond coding now - what's new

Quick Overview

Key Points

Outline

Similar Articles

Introducing GPT-5.3-Codex

OpenAI's GPT-5.3-Codex and Anthropic's Opus 4.6 Debut Amidst AI Rivalry

GitHub Copilot adds GPT-5.3-Codex with 25% faster performance for agentic tasks