OpenAI launches GPT-5.3-Codex, its most advanced coding model

OpenAI has presented the model GPT-5.3-Codexits most advanced coding artificial intelligence (AI) agent, capable of helping the company with its own creation.

The GPT-5.3-Codex model optimizes the encoding performance of GPT-5.2-Codex by also adding the reasoning and professional knowledge capabilities of GPT-5.2, and is overall 25 percent faster.

OpenAI has called GPT-5.3-Codex a “new industry milestone” in SWE-Bench Pro and Terminal-Bench, reference benchmarks used to measure coding, agent and real-world capabilities, as stated in a statement.

This model allows create complex video games and applications from scratch in a matter of days. To test it, OpenAI asked GPT-5.3-Codex to create two video games: a more advanced version of the racing game released with Codex and a diving game. Using web game development skills and generic “improve game” commands, the model was able to generate the video games autonomously.

On the other hand, GPT-5.3-Codex is designed to assist in the software life cycle with debugging, implementation, monitoring, PRD writing, copy editing, user research, testing, metrics, among others. Beyond coding, the model is also capable of create web pages, make presentations, spreadsheets or other tasksseeking to be a unique and general purpose agent.

Additionally, GPT-5.3-Codex allows users correct or add indications while the agent is working in real time. In this way, the model explains what it is doing so that the user can guide and adjust it more precisely.

ABLE TO HELP YOUR OWN CREATION

However, the main novelty of this model is that it has been the first to help OpenAI with its own creation. “The Codex team used early versions to debug his own trainingmanage its implementation and evaluate the results” the company explained on its website.

GPT-5.3-Codex is now available in all ChatGPT payment models in countries where it can be used Codex.