GPT-5.3-Codex is the new programming model of OpenAI, an agent powered by artificial intelligence that surpasses its predecessors in terms of performance and has contributed to its own creation and development.
This novelty improves the level of GPT-5.2-Codex and optimizes both its coding performance and speed, since It is 25% faster than this other oneas explained by the company in a statement.
In this he has also said that, thanks to this, the model can take on long-term tasks that «involve research, use of tools and complex execution.»
However, the most striking thing about this launch is that this model, which can write and review code, as well as do almost everything that a professional can execute on a computer, has contributed to its own development.
Develop websites and games with fewer tokens
The company led by Sam Altman has not spared compliments on its latest model, which it has said establishes a new industry record in parameters such as SWE-Bench Pro and Terminal-Bench, in addition to showing solid performance in OSWorld and GDPval.
With this, he recalled that SWE-Bench Pro covers four programming languages and is more resistant to pollution, diverse and relevant to the industry. This means that GPT-5.3-Codex breaks new records and does so with fewer tokens than any previous model.
To create highly functional games, applications and websites from scratch, you need a few days, and you do it with minimal instructions, as explained by OpenAI, which has advanced that it has tested this model with two games.
On the one hand, he was asked to generate a diving game and the second version of a racing game. For this, a couple of generic suggestions, such as «fix the bug» and «improve the game». With that alone, this new Codex model autonomously iterated games over millions of tokens.
Stronger results with simple instructions
To highlight the virtues of GPT-5.3-Codex, OpenAI has compared it with its predecessor, GPT-5.2-Codex. To both You were asked to create two web pages. In that case, the new agent’s response was more complete because it automatically displayed an annual discounted pricing plan and created a review carousel.
In this sense, OpenAI has stressed that this model is designed to support all the work of the software life cycle, that is, debugging, implementation, monitoring, text editing, testing and metrics.
In that way, this language can help developers create whatever they want, both presentations and data analysis in spreadsheets.
GPT-5.3-Codex is autonomous, but needs supervision
As is common in OpenAI and other companies dedicated to the creation of AI models, in addition to the great virtues of this model, it has been insisted that humans continue to be essential for this format to work.
In this sense, the technology company has stressed that the Codex application facilitates the management and direction of agents and that the model communicates more and better with developers.
Thus, it is able to provide process status updates so that programmers can stay «in the know» and know what their process is, rather than waiting for a final result.
GPT-5.3-Codex «can interact in real time, ask questions, discuss approaches and move towards the solution, explaining what you are doing, responding to comments and keeping the user informed from start to finish,» he concluded.