Codex comes with your ChatGPT plan, and so do its limits. This guide explains how they work on each plan, how to see what's left, what happens when you run out, and how to make a plan go further. It's all from OpenAI's own pages, checked on October 21, 2026.
(I build Morphy, which runs Codex next to Claude Code and other AIs. It gets one short section near the end.)
How do Codex's usage limits work?
Codex counts your usage in 5-hour windows, and weekly limits may also apply. How far a window goes depends on the model, because a message to a bigger model uses more of it.
| Plan | Codex | Local messages per 5 hours |
|---|---|---|
| Free and Go ($8) | In the desktop app only, with GPT-6 Luna (as it rolls out) | Not published |
| Plus ($20) and Business | The app, CLI, IDE extension and cloud tasks, with every model | GPT-6 Astra 5–45, GPT-6.1 Sol 15–160, GPT-6 Sol 15–150, GPT-6 Luna 350–3,000 |
| Pro ($100, $200 or $500) | Everything in Plus | No 5-hour limit; weekly limits may apply |
The ranges are wide because a message's size varies: a long task with a big codebase uses more than a quick question. Cloud tasks can use more of your allowance than local messages.
On the $200 Pro tier, new subscriptions get a lower allowance than before; subscriptions that were active between September 22 and 29 keep the old one until October 29, 2026.
How do you check how much you have left?
- The usage page at chatgpt.com/codex/settings/usage shows your current limits and when they reset.
- In the Codex CLI, type
/statusduring a session.
What happens when you hit the limit?
The agent can finish the turn it's working on, within fair use. After that, you can:
- Wait for the reset. The usage page says when.
- Buy credits. On Plus and Pro you can keep going with credits instead of upgrading.
- Switch to a smaller model. GPT-6 Luna gets about twenty times as many messages per window as GPT-6.1 Sol on Plus, which is plenty for routine work.
- Upgrade to Pro, which has no 5-hour limit.
- Use an API key instead, billed per token at API prices. It works in the CLI and the IDE extension, without the cloud features like GitHub code reviews.
- Hand the task to another AI. If you also have a Claude plan, Claude Code reads the same
AGENTS.md. See Codex vs Claude Code.
Credits and speed modes
Credits pay for usage beyond your plan. They're priced per million tokens, and the model makes a big difference:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | 250 credits | 25 | 1,250 |
| GPT-6.1 Sol | 50 credits | 2.5 | 250 |
| GPT-6 Luna | 2.5 credits | 0.25 | 12.5 |
So GPT-6.1 Sol costs a fifth of GPT-6 Astra, and GPT-6 Luna a twentieth of Sol. Repeated context (cached input) costs a tenth or less.
Speed modes cost more. Fast uses your plan's allowance two and a half times as quickly (twice as many credits). Ultrafast, only on the $500 Pro tier, uses it eight times as quickly (six times the credits).
8 ways to make your plan last longer
- Match the model to the job. GPT-6.1 Sol for most work, GPT-6 Astra only for the hard parts, GPT-6 Luna for quick, focused tasks.
- Stay on Standard speed unless you're waiting on a result.
- Lower the reasoning level for simple tasks. Levels run from Light to Ultra; Max thinks longer about one task, and Ultra uses subagents, so both use more.
- Work locally rather than in the cloud when you can: cloud tasks can use more of your allowance.
- Start a new chat for unrelated work, so each message doesn't carry a long history.
- Keep
AGENTS.mdshort and specific. It's read in every session. - Be specific. "Add input validation to the login form in auth.ts" reads a few files; "improve this codebase" reads everything.
- Run fewer agents at once, or spread them across plans: several Codex chats use your plan several times as fast.
How does it compare with Claude Code?
Both use 5-hour windows plus weekly limits, but they're shaped differently: Codex publishes message counts per model, and Pro drops the 5-hour limit. Claude Code shares its limits with Claude chat, and Max multiplies Pro's allowance five or twenty times. See Claude Code's usage limits and Codex vs Claude Code for prices side by side.
In Morphy
Morphy runs Codex with your ChatGPT sign-in, next to Claude Code, Copilot, Gemini, Grok and others. Settings → Plan usage shows how much of your Codex plan you've used and when it resets. When a chat hits the limit, it offers Send when it resets (on Pro) and Continue with Claude, or any other AI you've connected: a new chat beside it, in the same folder, that carries the conversation on. See Connect your AI.
Questions
How many messages do I get on ChatGPT Plus? Per 5-hour window, about 15 to 160 with GPT-6.1 Sol, 5 to 45 with GPT-6 Astra and 350 to 3,000 with GPT-6 Luna, with weekly limits possible on top.
Does ChatGPT Pro have unlimited Codex? No 5-hour limit, but weekly limits may still apply.
Can I use Codex for free? Yes: ChatGPT Free and Go include Codex in the desktop app, with GPT-6 Luna as it rolls out. The CLI, IDE extension and cloud tasks start at Plus.
