Addendum to o3 and o4-mini system card: Codex

OpenAI has published an addendum to the system card for its o3 and o4-mini models, detailing a new cloud-based coding agent called Codex. According to the OpenAI blog, Codex is powered by codex-1, a version of o3 specifically optimized for software engineering tasks. The model was trained using reinforcement learning on real-world coding challenges across diverse environments. This training enables codex-1 to generate code that closely aligns with human coding style and adheres to project-specific preferences, such as pull request conventions. A key feature is its ability to iteratively run tests until passing results are achieved, improving reliability. The addendum provides transparency into the model's capabilities and training methodology, which is valuable for developers evaluating AI tools for software engineering workflows. By releasing this information, OpenAI aims to help users understand how Codex differs from general-purpose models and when to apply it for tasks like automated code generation, testing, and refactoring. For developers building AI-powered development pipelines, this clarifies a specialized tool that can integrate into cloud-based IDEs or CI/CD processes.