Codex usage limits come from your ChatGPT plan. Sign in to Codex with ChatGPT and your local messages and cloud tasks draw on one allowance. On Plus it is counted in a five-hour window, with a weekly limit on top. On Pro it is counted by the week alone.
OpenAI gives no fixed message count. It gives ranges instead. On Plus, GPT-6.1 Sol allows roughly 15 to 160 local messages per five-hour window, and the lighter GPT-6 Luna 350 to 3,000. Where a task lands in that range depends on the model, the reasoning effort, the context and the tools it uses.
The limits also moved a lot this fall. OpenAI doubled the allowance on two Pro plans on October 29, and ran a 28-day campaign in October that handed out resets on days it had no feature to ship. Every number below comes from OpenAI’s own pages, linked as we go and read on November 3, 2026.
We make Context Board, a live board that keeps your Codex limits on screen, and it gets one short section near the end.
- Plus ($20): a five-hour window, plus weekly limits that may also apply. OpenAI estimates 15 to 160 local messages per window with GPT-6.1 Sol.
- Pro ($100, $200 or $500): no five-hour limit right now, only a weekly allowance. Pro 100 and Pro 500 got twice their previous allowance on October 29, 2026.
- One pool: Codex shares its allowance with ChatGPT Work, ChatGPT for Excel and Workspace Agents. ChatGPT’s image, voice and upload limits stay separate.
- Resets: a new five-hour window starts with your first message after the last one ends. Banked, automatic and bought resets restart a window and move its next refresh.
- Hit one? The current turn finishes. Then buy credits, use a reset, upgrade, wait, or switch to an API key. Context Board shows what is left before you get there.
Codex usage limits by plan
OpenAI’s Codex pricing page says the number of messages you can send “depends on the model used, size and complexity of your tasks, and whether you run them locally or in the cloud.” This is how each plan is set up as of November 3, 2026.
| Plan | Price | Five-hour limit | Longer limit | Codex Cloud |
|---|---|---|---|---|
| Free | $0 | Not published | Not published | No |
| Go | $8 a month | Not published | A monthly allowance | No |
| Plus | $20 a month | Yes, estimated per model | Weekly limits may also apply | Yes |
| Pro 100, 200, 500 | $100, $200 or $500 a month | None currently | Weekly allowance, larger on each tier | Yes |
| Business Standard | $25 a seat monthly, or $20 billed yearly, 2+ seats | Yes, same estimates as Plus | Weekly limits may also apply | Yes |
| Business Premium | $125 a seat monthly, or $100 billed yearly | None | Higher usage than Standard | Yes |
| Enterprise and Edu | Contact sales | No fixed rate limits on flexible pricing. Otherwise the same per-seat limits as Plus for most features | If the workspace turns it on | |
| API key | API rates per token | No plan allowance. You pay for what you use | No | |
Some notes on the rows.
- Free and Go: both get GPT-6 Luna at Standard speed in the desktop app, subject to rollout. OpenAI publishes no numbers for either. Its resets article describes Go’s main allowance as monthly.
- Codex Plus limits: Plus covers Codex on the web, in the CLI, the IDE extension and on iOS, plus automatic code review and Slack. OpenAI’s tagline for it is “a few focused coding sessions each week.” Plus usage did not change in the October 29 increase.
- Codex Pro limits: “Pro plans currently have no five-hour limit.” Pro 200 includes more than Pro 100, and Pro 500 has the most. Among Pro plans, Ultrafast mode comes only with Pro 500.
- Business: a Premium seat has no five-hour limit and higher Work and Codex usage than a Standard seat. Both can buy workspace credits to go past their limits.
- Enterprise on token billing: some Enterprise agreements bill Codex in US dollars instead of credits. Your admin sets the budget and limits there.
The five-hour estimates per model, for Plus and Standard Business, come from the same pricing page:
| Model | Local messages per five hours |
|---|---|
| GPT-6 Astra | 5 to 45 |
| GPT-6.1 Sol | 15 to 160 |
| GPT-6 Sol | 15 to 150 |
| GPT-6.1 Terra | 35 to 285 |
| GPT-6 Luna | 350 to 3,000 |
OpenAI says plainly that these “are not fixed message limits.” The spread inside each range is the task. A one-file fix sits at the top. A long session holding a big repo in context sits at the bottom.
What uses up your Codex limits faster
OpenAI lists the inputs: “Model choice, context, reasoning, tool use, retrieval, and caching all affect usage, so prompt length alone isn’t a reliable estimate.” These are the ones with a number or a clear rule attached.
- The model. Astra’s range on Plus is 5 to 45 messages. Luna’s is 350 to 3,000. Same window, same plan.
- Reasoning effort. OpenAI’s Astra usage guide says higher effort “can use more of your allowance and does not always produce a better result.” It also says Astra at Low can outperform Sol at High.
- Speed modes. Fast mode uses your included limits at 2.5 times the Standard rate. Ultrafast on GPT-6 Astra or GPT-6.1 Sol uses them at 8 times.
- Cloud over local. Cloud tasks may use more of your allowance than local messages.
- Images. Image generation uses included limits about 3 to 5 times faster on average.
- Context you carry. Every MCP server adds context to your messages, and so does a long AGENTS.md.
Some things do not count. With a ChatGPT sign-in, auto-review safety checks are free and stay outside your plan’s limits. Composer predictions on Pro don’t count during the beta, though a prediction you accept and send does.
Code review has its own rule. OpenAI counts “Code Review usage” only when Codex reviews through GitHub, for example when you tag @Codex on a pull request. A review you run locally with /review counts toward your general limits.
When Codex limits reset
The five-hour window does not roll minute by minute. OpenAI’s Astra guide says “a new window starts when you send your first message in Work or Codex after the previous window ends.” So if you send nothing for a day, the clock starts again with your next message. You can also hit the five-hour limit well before five hours pass, with plenty of week left.
The week refreshes on its own schedule. Your usage settings show the next refresh time, and OpenAI says to read it there “rather than calculating it” yourself, because resets move it.
There are three kinds of reset on top of the regular refresh.
- A banked reset: a free reset saved to your account until it expires. On Plus and Business Standard it restores the five-hour and weekly allowances, on Pro the weekly one, on Go the monthly one. Use it from Usage settings, or with
/usageand Redeem reset in the CLI. It only works once something it covers is under 80% remaining. - An automatic reset: OpenAI applies it to the accounts in an offer, and you don’t redeem it. OpenAI notes that a “global” reset “does not mean every account or every usage limit.”
- A bought instant reset: offered to selected Plus and Pro personal accounts. It restores the five-hour and weekly allowances at once, even if you still have some left, and it cannot be saved.
A reset restarts your week. Per OpenAI’s paid reset article, the new weekly period starts with your first request after the reset, and the next automatic reset comes seven days after that. A bought reset pulls next week’s allowance forward rather than adding extra, so buy one when the week is nearly empty and the refresh is still days away.
OpenAI Support does not reset limits by hand.
Why Codex limits kept resetting in October
If your limits refreshed out of nowhere last month, that was on purpose. On October 5, after DevDay, OpenAI started “28 days of Quality of Life improvements and/or a full Reset.” Each day brought one improvement for most Codex and Work users, and days without one brought a full reset instead. The OpenAI Developer Community thread kept the running list, with a link to each day’s post on X.
Resets landed on many of those days. The run closed at the start of November with banked resets for everyone, which you can keep until they expire. Check your usage settings for any still waiting.
It was not the first round. OpenAI’s Astra guide lists banked resets on September 3 and 4, 2026, and an automatic reset on September 7, to cover the Astra launch delay. Reset banking itself arrived for Plus and Pro on June 11, 2026, per the ChatGPT release notes.
Two more changes landed at the end of October:
- Doubled Pro limits: on October 29, OpenAI gave Pro 100 and Pro 500 twice their previous Work and Codex allowance. It also canceled a Pro 200 cut announced on September 29, and Pro 200 subscribers on the lower allowance were doubled to match the higher one.
- A reported December merge: the same day, Tibor Blaho posted a hidden checkout line reading “Starting in December, plan usage will have a shared limit across all of ChatGPT and Codex.” OpenAI’s help pages say nothing of it as of November 3, so treat it as unconfirmed.
If you saw “2x Codex rate limits” elsewhere, that was a separate limited-time promotion for Plus and Pro when the Codex app launched on February 2, 2026.
What happens when you hit a Codex limit
The turn you are in gets to finish. OpenAI says “the agent will be able to continue working on that turn, subject to fair use limits.” The next message is the one that stops.
Then the limit banner and the usage page show your options. Per OpenAI’s Codex with your ChatGPT plan article:
- Buy credits. Plus and Pro users can keep working without changing plan. Your included usage goes first, then the credit balance. Credits are valid for 12 months from purchase, per the credits article. Business, Edu and Enterprise plans on flexible pricing buy workspace credits.
- Use a reset. A banked one if you have one, or a bought instant reset if your account is offered one.
- Upgrade. Pro drops the five-hour limit, and Pro 200 and Pro 500 raise the week.
- Wait. The banner shows the reset time.
- Switch to an API key. Any user can run extra local chats with an API key, charged at standard API rates.
Switching models won’t help. The Astra guide says “switching models does not restore allowance in a shared usage pool.”
How to check where you stand
Type /status in a Codex CLI session to see your remaining limits. On the web, the Codex usage page shows which allowance is low, your credit balance and each reset time. Our guide to how to check Codex usage covers both, plus why platform.openai.com/usage shows API spend instead.
How to make your Codex limits last longer
These come from OpenAI’s own tips on the pricing page and in the Astra guide.
- Pick the smaller model for small jobs: Luna or Terra for routine edits, Astra or Sol when the problem is hard.
- Start at low or medium effort: raise it only when a result misses something.
- Leave Fast mode off unless you are waiting on it: 2.5 times the usage is a lot to pay for speed.
- Trim what rides along: precise instructions, only the files that matter, and an AGENTS.md nested by folder rather than one long file.
- Disable MCP servers you aren’t using: each one adds context to every message.
- Run small tasks locally: cloud tasks may use more.
My pick is the first two. They cost nothing, and the model table above shows Astra at the top end and Luna at the bottom are about 70 times apart. Then check the dashboard every week or two, as OpenAI suggests, to learn your pace.
If you also run Claude Code, its limits work a different way, shared with ordinary Claude chat. Claude usage limits covers that side, and Claude Code vs Codex compares the two plans head to head.
Keep your Codex limits in sight
/status and the usage page both wait for you to look. A board shows it while you work.
Context Board is ours, so weigh the bias. Its AI usage widget shows what is left of your Codex plan as rings that empty as you work: one for the current five-hour window, one for the week. They are green at half or more, amber from 20% and red under 20%, and each says when it resets, like “resets in 1h 5m” or “resets Sat 11:00 AM”. Claude Code sits beside it in the same widget.
Codex at 14% for the week. The big refactor waits for Saturday.
You can set an alert on the widget, at 10, 20, 30 or 50% left, or off. It starts at 20%. When a window drops below your line, its ring turns red and a notice shows on the board while it is open, once each time a window drops below.
The numbers come from Codex itself. As it works, Codex saves what OpenAI last told it about your limits in its session files. The runner, a small program you set up once with one line from Connect, reads the newest one and sends the board the percentages and reset times. It draws the windows Codex reports, so on Pro you see the week. From your sessions, only names, states and percentages leave your computer.
You add AI usage from the gallery, or start from the Agents template, which has it already.
Want one reading first, with nothing to install? The free Codex usage checker reads the newest limits line from the sessions folder you pick, in your browser, and uploads nothing. It reads once. Context Board is free for 3 days, then $12 a month, and the Codex usage page walks through the setup.