Tools for
Conviviality

Home / Benchmarks / OpenAI Codex

OpenAI Codex

OpenAI's coding agent for the terminal, with an open-source command line tool that works best with a ChatGPT login.

Benchmark review. Giants are reviewed rarely, to show what the tools in the directory are measured against. This one is not listed in the directory, and it is an example, not a recommendation. Scores are our opinion on the six questions, based on public sources and dated October 2026.

XL · SystemBenchmarkOpen source CLIHosted models
github.com/openai/codex
Preview image for OpenAI Codex

Image from github.com/openai/codex, taken 2026-10-30, shown to help you recognise the tool. It belongs to its maker. Ask us to remove it.

Codex is the odd one among the giants, because the command line tool is Apache-2.0 licensed and sits in a public repository where you can read it. OpenAI's README calls it a coding agent that runs locally on your computer, with builds for Mac, Linux and Windows and installs through npm or Homebrew. The page recommends signing in with a ChatGPT Plus, Pro, Business, Edu or Enterprise plan, and offers an API key as the other route, with extra setup.

The code is open and the models are not, so the agent does its thinking on OpenAI's servers, and the page does not say whether local models are supported. Power is a 2 because the open tool still leans on one company's accounts and models. We read the repository page and nothing else. As of October 2026 this is our opinion, based on public sources, and if you spot an error, write to hello@toolsforconviviality.org and we will check it and correct it. Compare it with the tools in Coding agents.

    What stands out

  • An open-source terminal coding agent from OpenAI, Apache-2.0
  • Recommends signing in with a ChatGPT plan, with an API key as the other route
  • The page does not say whether local models are supported