Codex pricing starts at $0, because every ChatGPT plan now includes Codex: Free, the $8 Go plan, Plus at $20 a month and Pro from $100. Paying more buys you more places to run it and more work per week, and if you would rather skip the subscription, an API key bills Codex per token instead.
Figures below come from OpenAI's Codex pricing page and its help center, checked on November 4, 2026. OpenAI edits these pages often, so treat them as a snapshot.
Codex pricing by plan
| Plan | Price | Codex on this plan |
|---|---|---|
| Free | $0 | GPT-6 Luna in the desktop app, subject to rollout |
| Go | $8 a month | The same desktop-app entry as Free |
| Plus | $20 a month | Web, CLI, IDE extension and iOS; a five-hour window and weekly limits |
| Pro 100, 200, 500 | $100, $200 or $500 a month, billed monthly | No five-hour limit for now, a larger allowance at each tier, Ultrafast on Pro 500 only |
| Business | $25 a user monthly, or $20 billed yearly, two users minimum | SSO, MFA, larger cloud machines |
| Enterprise, Edu | Contact sales | Admin and compliance controls |
| API key | Per token at API rates | CLI, SDK and IDE extension, no cloud features such as GitHub review or Slack |
What changed recently
- Pro 500 joined Pro 100 and Pro 200. Pro 100 and Pro 500 now carry twice their earlier Codex and Work allowance. Pro 200 is open to new subscribers again, and OpenAI dropped an allowance cut it had announced in a September 29 email.
- GPT-5.5 left Codex for ChatGPT sign-ins on October 14. API-key sessions kept it.
- Free and Go are narrower than they sound. UsageAtlas's own October guide listed both as included with lower limits. OpenAI's page now names one model in the desktop app, and lists the CLI from Plus upward.
What eats the allowance
OpenAI does not sell Codex by the message. It publishes rough estimates: on Plus, GPT-6.1 Sol fits about 15 to 160 local messages per five hours, and the lighter GPT-6 Luna fits 350 to 3,000. OpenAI says these are not fixed limits. The range is wide because a one-line fix and a big refactor each count as one message.
Your allowance shrinks faster with:
- larger models
- Fast mode, which spends it at 2.5 times the Standard rate, and Ultrafast at 8 times
- cloud tasks, which share one pool with local chats
- a long AGENTS.md file or many MCP servers, since both add context to every message
When you run out
The turn already running can finish. After that, Plus and Pro users can buy ChatGPT credits instead of upgrading, and anyone can switch to an API key at standard API rates. Credits are priced per million tokens for each model and expire 12 months after purchase; the Codex credits guide explains them.
One detail trips people up. Pro includes $100 to $500 a month in OpenAI API credits, but OpenAI says those do not pay for Codex through an API key and do not raise your Codex limits.
Price your own week first
A price table cannot pick your tier. OpenAI's usage dashboard and the /status command show where you stand right now, but not how last month went.
UsageAtlas fills that gap. It is a free, open-source desktop tracker for Windows, macOS and Linux that connects through the Codex CLI already signed in on your computer and shows:
- the session and weekly windows, with reset times
- your credit balance, where Codex reports it
- tokens, requests and models from local sessions, with an estimated cost
- a history that grows week by week
Local tracking needs no account, and UsageAtlas never reads prompts or responses. Its dollar figures are estimates, so OpenAI's billing page stays the record. Two weeks of history settle it: if Plus stops you once a month, credits are cheaper, and if it stops you most weeks, try Pro 100.
When a tracker is not the answer
Companies on Business or Enterprise get organization totals from the admin console. UsageAtlas covers your own machine, and it also tracks Claude Code, Cursor and other tools if you split work across them.