o3 and o4-mini System Card Addendum: Codex
Key point
OpenAI optimized the o3 model to unveil **Codex**, a coding agent that writes and tests code in a development environment.
Details
Codex is a cloud-based coding agent that runs on codex-1, a variant of OpenAI's o3 model optimized for software engineering. This model was trained with reinforcement learning on real-world coding tasks across diverse environments, and is designed to mimic human coding style and PR (Pull Request) preferences, follow instructions precisely, and iterate until tests pass.
Each agent runs in an isolated cloud container with internet access disabled. The container is preloaded with a development environment containing the code, dependencies, configuration, and tools specified by the user, and once set up, the model can read and edit files as well as directly run commands such as tests, linters, and type checkers.
To increase the reliability of its work, Codex cites terminal logs and files to provide verifiable evidence of execution results. Once a task is complete, users can review the results or request revisions, and can turn the generated diff into a GitHub Pull Request or export it to a local environment.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.