Competition in the AI segment continues to intensify. This week, Anthropic used the Super Bowl to launch an advertising campaign targeting the arrival of ads on ChatGPT . Sam Altman responded to the campaign by labeling the ads “clearly dishonest.” In a new development, OpenAI and Anthropic released their most advanced coding models within minutes of each other. Originally, both companies scheduled their big reveals for 10:00 a.m. PST. However, in a strategic move to grab the spotlight first, Anthropic moved its release forward by fifteen minutes. OpenAI responded immediately, unveiling GPT-5.3-Codex.
This new GPT-5.3-Codex model represents more than a simple speed boost. OpenAI claims that it marks a transition from a tool that merely “writes and reviews code” to an autonomous agent capable of handling nearly any task a professional does on a computer.
OpenAI’s GPT-5.3-Codex: 25% faster and agentic AI
The most striking detail about GPT-5.3-Codex is its role in its own creation. OpenAI revealed that the development team used early versions of the model to debug training runs, manage deployments, and diagnose test results. Essentially, the AI helped build its own final version.
While this sparks discussions about “technological singularity”—the point where technology improves itself without human intervention—for now, it simply means a much more efficient development cycle. This self-acceleration allowed the team to produce a model that is 25% faster than its predecessor, GPT-5.2.
There is a lot more on software engineering than just writing code. Engineers’ task list includes research, fix bugs, keep an eye on systems, and handle deployments. The goal of GPT-5.3-Codex is to help with this whole process. The model built functional games and complex apps from scratch in just a few days during internal testing.
The model also sets new standards for cybersecurity . It is the first OpenAI model to be rated “High capability” for security tasks, and it was specifically trained to find software flaws. OpenAI is giving $10 million in API credits to help defenders find and fix bugs before they can be used with malicious goals.
Collaboration in real time
Unlike previous versions where users would wait for a final output, GPT-5.3-Codex works like a true collaborator. It provides frequent updates on its decisions, allowing users to ask questions or steer the solution while the work is in progress.
The competition remains fierce, with Anthropic’s Claude Opus 4.6 also hitting the market simultaneously. Both companies report that their own engineers now perform the vast majority of their daily coding tasks using these AI agents.
GPT-5.3-Codex is now available for users on paid ChatGPT plans across desktop apps, CLI, and IDE extensions. These advances turn AI into an active partner that can operate a computer end-to-end.
