GPT-5.3-Codex explained: What OpenAI’s latest coding model can do

OpenAI has launched a new AI model called GPT-5.3-Codex, aimed at handling agent-style coding and software development work. Compared to earlier Codex versions, the company says the model is built to take on longer and more involved tasks, instead of being limited to writing or checking small pieces of code. The details were shared by OpenAI in an official announcement, and further reported by outlets covering the release. Also Read: Anthropic’s Claude Opus 4.6 AI model brings bigger context and smarter reasoning

What GPT-5.3-Codex Is Designed For

GPT-5.3-Codex is designed to support the entire software development process. Rather than stopping at writing code, it can also help with researching requirements, fixing issues in large codebases, using tools, and rolling out changes. OpenAI says the model can stay active across long-running tasks without losing context, allowing users to step in mid-task to ask questions or suggest changes. Also Read: From Resumes to Interviews: ChatGPT prompts that can boost your job search

According to OpenAI’s post, this version combines the coding performance of GPT-5.2-Codex with the reasoning and domain knowledge of GPT-5.2 into a single system. The company also claims it runs around 25 percent faster than previous Codex models, which helps when dealing with larger projects.

A Model That Helped Build Itself

One of the more notable points mentioned by OpenAI is that GPT-5.3-Codex assisted in its own development. Early versions of the model were reportedly used by the Codex team to debug training runs, manage deployment steps, and analyse evaluation results. OpenAI said this helped speed up internal development, marking the first time one of its models played a direct role in building itself.

Performance and Benchmarks

OpenAI shared internal benchmark results to show how GPT-5.3-Codex compares with earlier models. On SWE-Bench Pro, a test focused on real-world software engineering tasks, the model scored slightly higher than GPT-5.2-Codex. It also showed gains on Terminal-Bench and OSWorld-Verified, which measure performance on tool use and desktop-based tasks.

The company says the model can build full applications from basic prompts, including complex web-based games and websites, and iterate on them over long sessions.

Availability and Safety

GPT-5.3-Codex is currently available to users on paid ChatGPT plans across web, mobile apps, command-line tools, and IDE extensions. OpenAI has said API access will be added later.

On the safety side, OpenAI noted that GPT-5.3-Codex falls under its higher capability category for cybersecurity-related tasks and includes additional monitoring and access controls.

Add Techlusive as a Preferred Source

The release comes as competition in AI coding tools continues to intensify, with other companies also pushing new models aimed at long-running, agent-style workflows.