5>CLI agent· 5 models rankedOpenAI only

Best Models for OpenAI Codex

OpenAI's coding agent — ranked across the GPT-5 model lineup

Codex is OpenAI's agentic coding tool, and like Claude Code it is locked to one vendor's models — the GPT-5 family. Choosing a model means trading off frontier capability, the Codex-specialized checkpoints, and faster/cheaper tiers for high-volume turns. GPT-5.5 is the recommended default; the Codex-tuned 5.3 and the mini tier round out the lineup. (OpenAI publishes per-model token pricing on its pricing console rather than a static page, so exact $/MTok is linked rather than reproduced here — see Sources.)

Best pick

GPT-5.5Cloud API

OpenAI's newest frontier model and the recommended default for most Codex tasks — complex coding, computer use, and long agentic runs.

What Codex needs

Strong agentic tool use over long sessions, with a Codex-tuned checkpoint when you want coding-specialized behavior and a mini tier when latency and cost beat raw depth.

OpenAI models, ranked

GPT-5.5Cloud APIOpenAI · Flagship— / — $/MTok

OpenAI's newest frontier model and the recommended default for most Codex tasks — complex coding, computer use, and long agentic runs.

GPT-5.3-CodexCloud APIOpenAI · Balanced— / — $/MTok

Codex-specialized checkpoint tuned for terminal coding and tool use — a steady workhorse when you want coding-focused behavior.

GPT-5.4Cloud APIOpenAI · Prior gen— / — $/MTok

The prior general frontier model, still selectable as a reliable fallback to GPT-5.5.

GPT-5.4-miniCloud APIOpenAI · Fast— / — $/MTok

Faster, cheaper tier for lighter edits and high-volume turns where latency beats depth.

GPT-5.2Cloud APIOpenAI · Prior gen— / — $/MTok

Older selectable checkpoint, kept for compatibility and reproducibility; superseded by 5.4/5.5.

How to use it

Install the Codex CLI and sign in with ChatGPT or an API key. The model picker offers GPT-5.5, GPT-5.4, GPT-5.4-mini, GPT-5.3-Codex, and GPT-5.2; for ChatGPT-authenticated sessions the simplest path is to leave it on the default (GPT-5.5 where available, GPT-5.4 as fallback).

Codex official site ↗

New open-weight models, real Apple Silicon benchmarks, and the one model worth running on your Mac this week. Free, one email a week, unsubscribe anytime.

Frequently Asked Questions

Which model should I use in Codex?+

GPT-5.5 is the recommended default for most Codex work. Use GPT-5.3-Codex when you want the coding-specialized checkpoint, GPT-5.4 as a fallback, and GPT-5.4-mini for fast, cheap, high-volume turns. If you are signed in with ChatGPT, leaving the picker on default is usually best.

Can Codex run on non-OpenAI or local models?+

No — Codex is locked to OpenAI's GPT-5 family. If you want a terminal coding agent that runs local open-weight models, OpenCode and Open Claude Code in our tools list are designed for that.

What is GPT-5.3-Codex versus GPT-5.5?+

GPT-5.3-Codex is a checkpoint tuned specifically for coding and tool use, while GPT-5.5 is the newest general frontier model. Most users get the best results from GPT-5.5; the Codex-tuned model is a strong, predictable alternative for coding-heavy sessions.