Codex usage limits explained: 5-hour and weekly windows
On ChatGPT plans, Codex usage is counted in two windows: roughly five hours, and a week. As of late October 2026, Plus and standard Business seats have both; Pro plans currently have no five-hour limit, though weekly limits may apply. How much fits in a window depends mostly on the model and how much context each request carries. Check what’s left with /status in the CLI or on the usage dashboard, and when you run out, wait for the reset, use credits, or switch to an API key.
These limits change often. OpenAI has changed the five-hour limit several times in 2026. The numbers below are what OpenAI’s pages said when we checked; your usage dashboard is the source of truth for your account.
- How the windows work
- Limits by plan
- Check what’s left
- When you hit a limit
- What uses it up faster
- The 5-hour limit in 2026
- Questions
How the windows work
Codex counts what you use against two windows at once: one of about five hours and one of a week. Each has its own reset time, and whichever runs out first stops you. The CLI shows both, labelled something like “5h limit” and “weekly limit”. OpenAI doesn’t publish exactly how the rolling window is calculated, so treat the reset times on your dashboard as the reference. Since September 2026, the CLI warns Plus and Team users earlier, once less than half of a five-hour window remains.
Limits by plan
OpenAI gives estimates in local messages per five-hour window, not in tokens. For Plus and standard Business seats (Codex pricing):
| Model | Local messages per 5 hours |
|---|---|
| GPT-6 Luna | About 350–3,000 |
| GPT-6.1 Sol | About 15–160 |
| GPT-6 Sol | About 15–150 |
| GPT-6 Astra | About 5–45 |
- Pro: “currently no five-hour limit”, but weekly limits may apply. OpenAI offers several Pro tiers, and new subscriptions to one of them get a lower allowance than older ones (Pro plans).
- Enterprise and Edu with flexible pricing: no fixed rate limits; usage draws on credits. Without flexible pricing, seats get the same limits as Plus.
- Free and Go: OpenAI doesn’t publish numbers.
- Weekly limits: no numbers are published for any plan.
The ranges are wide because a “message” can be a one-line fix or an hour-long refactor that reads half the repository. Some of OpenAI’s own pages disagree with each other, for example a pricing page that still lists retired models; we went by the documentation page above.
How to check what’s left
- In the CLI:
/statusshows the current session and your 5-hour and weekly windows./statuslinecan keep them in the footer, and/usageshows your token activity. - On the web: the Codex usage dashboard shows limits and reset times.
- In the ChatGPT desktop app:
/usageopens usage and billing settings. - On iPhone: the ChatGPT app has Home Screen and Lock Screen widgets with what’s left and when it resets.
What happens when you hit a limit
A turn that’s already running can usually finish. After that, Codex stops with a message like “You’ve hit your usage limit” and the time you can try again. Your options (using Codex with your ChatGPT plan):
- Wait for the reset. It’s the cheapest option, and a reasonable time for a real break.
- Use credits on Plus and Pro, or workspace credits on business plans.
- Upgrade your plan.
- Switch to an API key, which is billed per token instead of by window.
- Use a banked reset if you have one. You earn them through referrals and promotions, and a full one refreshes both your 5-hour and weekly windows and moves your weekly reset date. You can’t buy one, and support can’t reset limits for you.
What uses it up faster
OpenAI lists model choice, context, reasoning, tool use, retrieval and caching as what drives usage. In practice:
- The model. As the table shows, the largest model fits dozens of times fewer messages than the lightest. Use a lighter one for routine edits.
- Reasoning effort. Higher effort means more tokens per answer.
- Long threads. Every request carries the conversation so far. Start a new thread for a new task, and compact long ones.
- Always-loaded context. Each connected MCP server and a large
AGENTS.mdadd to every request. Trim what you don’t use. - Speed modes. Against included usage, Fast counts about 2.5 times and Ultrafast about 8 times.
- Cloud tasks can use more than local ones, and image generation uses limits several times faster.
The five-hour limit in 2026
The five-hour window has come and gone this year. OpenAI support confirmed in July that it was lifted for Plus, Pro and Business on July 12, 2026. A return planned for the end of July was postponed, and in late August OpenAI’s Thibault Sottiaux said on X that it was coming back for Plus on August 25 while staying off for the main Pro tiers for the coming months. The CLI’s September release added the earlier warning for Plus and Team, and the current pricing page lists five-hour estimates for Plus. These dates come partly from posts rather than an official announcement, which is one more reason to trust your dashboard over any article, this one included.
Seeing the limits at a glance
We make Meowdoro, a focus timer around the MacBook notch that shows Codex and Claude Code sessions, so this is a disclosure. It shows Codex’s 5-hour and weekly limits as one small pie for whichever window is fuller. The pie turns orange from 80%, or earlier when, at your current pace, the window would run out before it resets, and it says when that would happen. Machines signed in to the same ChatGPT account share one group and its limits (how it shows agents). And if you hit the limit anyway, that’s a break; see what to do while you wait.
Questions
Does Codex have a 5-hour limit?
On Plus and standard Business seats, yes, as of late October 2026. Pro plans currently don’t, but weekly limits may apply. Check /status or your dashboard for your account.
When does the Codex weekly limit reset?
At a time shown on your usage dashboard and in /status. It’s tied to your account, not to a calendar day, and a banked reset moves it.
How do I check my Codex usage?
/status in the CLI, the usage dashboard at chatgpt.com/codex/settings/usage, /usage in the ChatGPT desktop app, or the widgets in the ChatGPT iPhone app.
Can I buy a limit reset?
No. You can buy credits on Plus and Pro, but resets are earned through referrals and promotions, and support can’t reset limits.
How do I get notified when Codex finishes, so I don’t waste the window waiting?
See Codex notifications: built-in CLI notifications, hooks and the ChatGPT app.
Sources: OpenAI’s Codex pricing documentation, help articles on using Codex with a ChatGPT plan and Pro plans, the Codex release notes, an OpenAI community forum post from OpenAI support, and a post on X by Thibault Sottiaux of OpenAI, read on October 26–27, 2026.