OpenAI has introduced a new model, GPT-5–Codex, now available to ChatGPT Plus, Pro, Business, Edu, and Enterprise users through Codex in the terminal, IDE, GitHub, or ChatGPT. This model is designed for programming automation and is aimed at developers, engineers, and scientists using AI for coding. The model allows for both short interactive sessions and long tasks, taking from a few seconds to seven hours, depending on the complexity.
GPT-5-Codex implements a new approach to task execution time distribution, dynamically determining how many resources are needed for each task. It can work on large projects, supports more programming languages, and confidently handles complex development stages. The model surpasses the previous version of GPT-5 on standard AI assistant tests, including SWE-bench Verified and code refactoring tasks from large repositories.
The model was separately trained for code review, and experienced programmers noted the quality of its comments. GPT-5-Codex leaves fewer incorrect remarks and more useful suggestions during the change verification process. It can check commits, dependencies, run tests, and compare expected changes with actual ones.
Users can access Codex via the command line, IDE extensions, or a cloud service where additional automation features are available. For example, the system installs missing dependencies, analyzes setup scripts, generates task lists, and can attach screenshots to pull requests. Developers with Business and Enterprise plans receive enhanced capabilities for working on multiple projects and can purchase additional usage capacity. OpenAI plans to add access to GPT-5-Codex via API soon.
