OpenAI Unveils GPT 5.3 Codex

OpenAI has released GPT 5.3 Codex, a powerful new artificial intelligence model designed for advanced agentic coding tasks. Announced on Thursday, the San Francisco based AI company said the model enables Codex to independently handle complex software workflows, including building full scale applications and video games from scratch. OpenAI described it as its most capable agentic coding system to date, capable of managing entire development lifecycles from research and debugging to deployment.

GPT 5.3 Codex is now available globally to all paid ChatGPT users across mobile and desktop apps, the command line interface, IDE extensions, and the web, with API access coming soon. The model combines the frontier coding strength of GPT 5.2 Codex with the deep reasoning and professional knowledge of GPT 5.2 into a unified system. It is 25 percent faster and supports long horizon projects involving research, tool usage, and complex execution, while allowing users to guide and adjust tasks in real time without losing context.

According to OpenAI’s internal benchmarks, GPT 5.3 Codex shows notable performance gains across multiple evaluations. It scored 56.8 percent on SWE-Bench Pro, 77.3 percent on Terminal-Bench 2.0, and 64.7 percent on OSWorld-Verified. The model can autonomously generate complex web games, production-ready websites, and supports the full software lifecycle beyond coding. OpenAI also confirmed that GPT 5.3 Codex played a role in its own development and is the first model classified as “High capability” for cybersecurity tasks under the company’s Preparedness Framework, backed by enhanced safety and monitoring systems.