OpenAI Unveils GPT-5 Codex for Advanced Programming Tasks
OpenAI has introduced GPT-5 Codex, a new model designed to independently tackle complex programming tasks and ensure software reliability. This model is now available in the Codex platform, integrated as the standard tool for cloud tasks and code reviews. It is included in various ChatGPT plans, offering multiple projects per week for higher tiers.
GPT-5 Codex is specifically trained for code reviews, analyzing commits, checking dependencies, running tests, and comparing intended changes with actual diffs. It can be used via the CLI, IDE extensions, and the Codex Cloud environment, creating to-do lists and accessing external tools. The cloud version can recognize common setup scripts, install missing dependencies, and start its own browser to check visual results.
In testing, GPT-5 Codex worked on complex tasks for over seven hours, adjusting compute time dynamically based on task complexity. It can operate in local development environments, the cloud, or directly on GitHub. An API connection for GPT-5 Codex is announced but not yet available.
GPT-5 Codex, a variant of GPT-5 optimized for real development tasks, outperforms its predecessor in standardized tests. It achieved 74.5% accuracy in SWE-bench Verified and 51.3% in refactoring tasks. With its integration into the Codex platform and various ChatGPT plans, GPT-5 Codex is set to become a valuable tool for developers.
