Directory / OpenAI Codex
OpenAI Codex
Verified
OpenAI’s coding agent across CLI, IDE, and cloud tasks.
Codex is OpenAI’s agentic coding product: a CLI and IDE surface that reads a repository, edits files, runs commands, and can continue work as a cloud task. It is locked to OpenAI models. The name recycles the 2021 Codex model; the 2025–2026 product is an agent harness, not an API model card.
- Model independence
- Locked
- Deployment
- Cloud, Self-hosted
- Open source
- No
- License
- Proprietary
- Pricing model
- Mixed
- Added / updated
- 19 Aug 2026 / 19 Aug 2026
Editorial take
Codex is what you buy when the factory is allowed to standardize on OpenAI the way Claude Code standardizes on Anthropic. The CLI is real and CI-usable. It is not model-agnostic and not open source. Treat cloud tasks as OpenAI-hosted computers, not as a box you inspect. Compare it to Claude Code, not to Aider.
Strengths
- First-party harness for OpenAI models, including a documented CLI.
- Cloud tasks cover unattended work the local CLI should not hold credentials for.
- Same vendor as the models many eval harnesses already call.
Limitations
- Model-locked to OpenAI. Hosting via Azure still serves OpenAI models.
- No public repository for the product.
- Local CLI execution is not a self-hosted control plane.
Compare
- vs claude-code — The two lab-shipped coding agents that lock you to a model family.