GPT-5.3-Codex: OpenAI introduces new coding model
OpenAI has released GPT-5.3-Codex, a new coding model that, according to the development team, was significantly involved in its own development.
(Image: Tada Images/Shutterstock.com)
OpenAI's GPT-5.3-Codex is released just under two months after the release of GPT-5.2-Codex, which was released in mid-December. According to developers, the new version combines the coding capabilities of GPT-5.2-Codex with the reasoning and knowledge capabilities of GPT-5.2. It is said to be 25 percent faster than its predecessor.
Helped with its development
OpenAI further states in the announcement that GPT-5.3-Codex achieves new best scores in two of the four benchmarks (SWE-Bench Pro, Terminal-Bench) in four performance tests for programming and practical applications, and shows strong performance in the other two (OSWorld, GDPval).
According to OpenAI, GPT‑5.3‑Codex is the company's first model to be “significantly involved in its development.” To achieve this, the Codex team used early versions “to debug its training, manage its deployment, and diagnose test results and evaluations” – thereby contributing to accelerating its development, it is stated.
Videos by heise
Furthermore, OpenAI explains that GPT-5.3-Codex surpasses its predecessor as well as the newly released Claude Opus 4.6 (65.4 percent) by about twelve percent in the agentic coding benchmark Terminal-Bench 2.0 – OpenAI's new tool achieved 77.3 percent in this benchmark. The model consumes fewer tokens than its Codex predecessors.
(Image:Â OpenAI)
OpenAI plans to make GPT-5.3-Codex available to paying ChatGPT users in all Codex environments: in the app, via the CLI, in the IDE, and on the web. OpenAI is also working on “enabling secure API access soon.” Additionally, Apple announced a few days ago that it would integrate AI coding agents like Claude and Codex directly into the development environment Xcode from version 26.3.
The development environment uses the open-source Model Context Protocol (MCP) for this purpose, which gives AI agents access to project structure, build logs, and Apple's documentation. The new features build on the integration of external Large Language Models (LLMs) into Xcode 26, which already enabled automatic transfer of source code to GPT and Claude without copy & paste. GitHub also announced the integration of the two coding agents.
(afl)
