OpenAI Codex is a cloud-based software engineering agent designed to handle multiple coding tasks in parallel, accessible via ChatGPT Pro, Team, and Enterprise
Codex operates in isolated cloud sandboxes, where it can read, edit, and test code, with each task tracked and verifiable through logs and test outputs
Users interact with Codex through the ChatGPT sidebar, assigning tasks by prompt and selecting “Code,” with real-time progress tracking
Codex can write features, fix bugs, refactor, write tests, scaffold new features, draft documentation, and submit pull requests
The agent uses AGENTS.md files and project documentation to better understand project structure, commands, and best practices
Codex-1, an optimized version of OpenAI o3, powers Codex, trained on real-world coding tasks for human-like code and PR style
Codex CLI is a lightweight, open-source terminal agent using a scaled-down codex-1 (o4-mini) for fast code editing and Q&A
Security is prioritized: Codex runs in isolated containers, blocks internet access during execution, and only uses code from GitHub and user dependencies
Pricing for developers is $1.50 per 1M input tokens and $6 per 1M output tokens, with prompt caching discounts; initial access is free for some users
Future plans include more interactive workflows, mid-task coaching, deeper tool integrations, and asynchronous agent collaboration