OpenAI has released GPT-5 Codex, an updated version of its GPT-5 model tailored for practical software engineering tasks. Unlike earlier versions, GPT-5 Codex is designed to handle complete software projects, implement features, debug, carry out large-scale code refactors, and perform code reviews.
The model adjusts its approach depending on task complexity, offering quick responses for small requests and sustained effort for larger assignments. In internal tests, it worked independently for over seven hours, resolving issues and delivering functional implementations.
GPT-5 Codex includes a code review feature intended to identify critical flaws more accurately. In evaluations using open-source commits, its review comments were assessed as more relevant and reliable, helping developers focus on key issues.
The tool is included with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans, with usage limits varying by plan. Enterprise customers have access to shared credit pools to support team use. GPT-5 Codex is accessible via command-line interface, IDE extensions, GitHub, the web, and the ChatGPT iOS app. Developers can switch between local and cloud environments without losing context.
The model operates in sandboxed environments with network restrictions. It requests permission before executing sensitive commands and supplies logs and test results for transparency. OpenAI has stated that Codex should serve as an additional reviewer, rather than replace human oversight.
The release is positioned as part of OpenAI’s broader aim to develop AI tools that assist in software development by adapting to context and improving reliability.
