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

OpenAI has rolled out a new coding-focused AI model, GPT-5.3-Codex, at a time when competition in developer-facing AI tools is heating up fast. The company says this is its most capable agentic coding model so far, designed not just to write code, but to handle long, complex tasks that stretch across the entire software workflow. The launch also comes on a busy week for OpenAI, which recently introduced a dedicated Codex app for macOS, hinting at a sharper focus on AI agents that work alongside developers rather than just answering prompts.

What makes this release interesting is not only the performance jump, but also OpenAI’s claim that GPT-5.3-Codex played a role in building itself, a statement that has already sparked curiosity and debate in the AI community.

A faster, smarter Codex

According to OpenAI, GPT-5.3-Codex brings together the strong coding abilities of GPT-5.2-Codex with the broader reasoning and professional knowledge seen in GPT-5.2. The result, the company claims, is a single model that is around 25 per cent faster than its predecessor. That speed boost matters because Codex is meant to run for long periods, handling tasks that involve research, tool usage, debugging, and execution without constantly needing restarts.

OpenAI describes GPT-5.3-Codex as a “frontier model,” a term commonly used in the AI industry for systems operating at the cutting edge of current capabilities. The faster runtime allows the model to take on longer-running jobs, including processes that can stretch beyond a full day.

An AI that helped build itself

The most eye-catching part of the announcement is OpenAI’s claim that GPT-5.3-Codex is the company’s first model that was “instrumental in creating itself.” In its blog post, OpenAI said the Codex team used early versions of the model to debug training runs, manage deployment, and analyse test results.

“The Codex team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations, our team was blown away by how much Codex was able to accelerate its own development,” OpenAI wrote.

This idea of AI systems assisting in their own development is not entirely new, but it is one of the clearest examples yet from a major AI lab. Rival Anthropic recently made a similar claim about its Claude Cowork system, and engineers across top AI companies now openly say that most of their day-to-day coding work is already handled by AI tools.

More than just writing code

OpenAI is positioning GPT-5.3-Codex as far more than a code generator. The company says the model can support almost every stage of the software lifecycle. That includes debugging, deployment, monitoring, writing product requirement documents, editing copy, user research, testing, and tracking metrics. It can also help with non-traditional developer tasks such as building slide decks and spreadsheets.

The model is designed to be steered mid-task, allowing users to interact with it while it works without losing context. This continuous interaction is expected to pair well with the new Codex Mac app, which acts as a control centre for managing multiple AI agents working in parallel.

Longer tasks, better intent

Another key upgrade is the ability to handle very long jobs. OpenAI says GPT-5.3-Codex can run processes that last more than a day. As part of internal testing, the company used Codex to build two web-based games, consuming millions of tokens over extended runs.

OpenAI also claims the model has a better understanding of user intent. For example, if a user gives a vague prompt to build a simple website, the model now defaults to adding sensible features and structure, offering a stronger starting point instead of a bare-bones result.