OpenAI has introduced GPT-5.3-Codex, a new version of Codex designed to work with code and digital tools. Compared with the previous model, it runs about 25% faster, resulting in quicker responses and less waiting time when handling complex tasks. The model combines code-writing capabilities with the ability to analyze problems and arrive at practical solutions.
The key difference between GPT-5.3-Codex and earlier Codex versions is that it is not limited to writing and reviewing code. The model can operate within a digital environment: opening and editing files, using the terminal, running programs, and interacting with web interfaces, while carrying tasks through from start to finish. This allows Codex to take on specific parts of professional workflows — from preparing code to related technical actions — while remaining under human supervision at any stage.
During the development of GPT-5.3-Codex, OpenAI’s team used the model itself as a working tool. It assisted in identifying errors, analyzing test results, and speeding up launch preparation. This helped improve the model more quickly and demonstrated that Codex can support not only end users but also AI developers themselves.
Alongside the model, OpenAI also unveiled Frontier, a business platform designed to bring AI agents into real-world company operations. Frontier provides agents with access to the necessary data and systems while enforcing clear rules and boundaries. As a result, organizations can use AI not as a standalone experiment, but as a digital assistant that genuinely saves time and supports everyday work.