Codex CLI
OpenAI's open-source terminal agent for OpenAI's own models. Released in 2025 as the company's answer to Claude Code, with a similar tool-use loop, sandboxed code execution, and a plug-in model that aligns with the OpenAI Apps SDK. The MIT-licensed source on GitHub is the reference for how OpenAI thinks about agentic coding harnesses.
Leaderboard Placements
Live board as of 2026-10-28: federated Terminal-Bench rows, with TensorFeed editorial scores for harnesses the board does not cover.
| Benchmark | Best base model | Score | Rank |
|---|---|---|---|
| SWE-bench Verified | - | - | - |
| Terminal-Bench | - | - | - |
| Terminal-Bench 2.1 | GPT-6 Astra | 87.4 | #1 / 12 |
| Terminal-Bench 3.0 | GPT-5.6 Sol | 34.6 | #1 / 11 |
| Terminal-Bench 4.0 | GPT-6 Astra | 58.2 | #3 / 21 |
| Aider Polyglot | - | - | - |
| SWE-Lancer | - | - | - |
Distribution
Open-source CLI. Install via npm or build from source. MIT license.
Model Story
OpenAI models only. The default follows OpenAI's current coding model, with cheaper or higher-reasoning models selectable per task.
Pricing
Free harness; you pay for the underlying OpenAI API tokens.
Who It's For
Engineers who want a terminal agent on the OpenAI stack and prefer a transparent, MIT-licensed implementation they can fork and extend.
Notable Features
- Sandboxed shell and code execution
- OpenAI Apps SDK plug-ins
- Open MIT license
- Auto-approve mode for unattended runs
- Native streaming tool-call UI