What does "Your dot is on a break" mean? Not your plan limit: how long it lasts and what to do, from 25 reports

The short version: what "on a break" means, what to do, and how to lose less work

What it is

"Your dot is on a break" is an abuse-prevention pause on your Dot itself, not your ChatGPT plan running out. The full message reads:

Your dot is on a break. Congrats you are in the very top users and have hit our abuse prevention limit. Check back in a bit!

People have hit it with 47% of a $200 plan left (c2chq), on the $100 plan (Genghis Khan II) and on Pro $500 (Dan Siegel). No meter shows how close you are, and the message gives no time. Where people timed it, the pause lasted from about 1 hour 30 minutes to about 13.5 hours. Writing again only brings the same message back, and nobody has reported that Reboot ended one early.

A second stop, "You've hit the preview limit. Try again at…", now often gives a retry date about a month away (13, 14 or 15 November in reports from 14 to 16 October). Treat that date as a guide rather than a lock: in two of those reports, some cloud-browser, video and Google Drive work ran again later the same day.

What to do, in order

1. Don't use Reset. It deletes the Dot with its conversations, memories and scheduled tasks. Reboot does no harm, but it hasn't ended a break for anyone.
2. Note when the message first appeared and your timezone. If you got a preview-limit date, write down that date and the time you saw it, because the message names no timezone.
3. Stop messaging the Dot. Come back later and send one short message to see whether it has returned.
4. Open Activity. Codex or Work tasks the Dot started can keep running or may already be finished, and they report there.
5. Check Settings → Usage. It won't show the break, but it shows whether your Work and Codex allowance is fine. Take a screenshot if a delegated task says you are out of usage while this page shows plenty left.
6. If something was being sent or uploaded when the stop hit, check whether it went out before you ask for it again. Look in Sent, or in the folder or page it was going to.
7. When the Dot is back, ask what is done, what isn't and where it stopped before you give it anything new.
8. If it lasts, email support@openai.com with a screenshot of the message, the affected task, when it stopped and your timezone. OpenAI Support asked for exactly that list in the community thread.

Where it gets stuck, and what helped

• The same message keeps coming back. Collin Lemmon told his Dot to keep going, Barna asked for a status update and StanBrusse asked "you there?", and all three got the break message again. Waiting is the only thing anyone has reported working.
• An overnight job stops without a word. c2chq's Dot stopped partway through filling a Google Sheet and never mentioned it. Ask for results to be saved as they are found, with the time of the last update at the top.
• The Dot comes back but its tools aren't ready. bshane54's Dot said it could reply but hadn't checked its background tools. Ask it to test the tools a job needs before it carries on.
• A send fails partway. dhruvashp's user-approved Gmail send was stopped by "Automatic approval review failed: You've hit the preview limit" before it ran, and the email was confirmed not sent, while reading mail and creating drafts still worked. noob000001 didn't repeat an upload whose result was unclear, to avoid a duplicate. Check first, then retry once.
• A delegated task says you're out of usage while the Usage page says you aren't. ozin-he saw it with 85% of the weekly window used. cveralyon's Dot, Exion, had a Codex Cloud task rejected before it started, with a retry date of 14 November, while Usage & limits showed 98% of the weekly agent allowance left and 3,162 credits. Open that task in Activity, screenshot both screens and send them to support, as cveralyon did.
• A retry date a month away. MrDias, AI驚かない屋, noob000001, Eta06 and Jeff Plaza's Dot each reported one. Try one small action later the same day before you plan around the date.
• Side-by-side tasks stop. With his parallel runs blocked, MrDias moved to running one task after another. Ask the Dot to queue work instead of starting several runs at once.
• Support replies don't help. Genghis Khan II found email answers generic. Put the screenshot, the task, the start time and your timezone in the first email.

What isn't possible or known yet

• There is no meter for the Dot's own allowance and no published threshold, so you can't see a break coming.
• Neither message gives a timezone, and OpenAI Support couldn't give a cooldown length in the community thread.
• Reboot, Reset and messaging again don't end a break. Reset makes things worse.
• Delegated Codex and Work tasks use those products' own allowance, so they can keep running while the Dot is paused, or fail on their own limit while the Dot is fine.
• People keep asking OpenAI for two things: a visible meter, and a predictable fallback such as a weekly limit instead of an open-ended pause.

One message to send before a long job

Before you start, split this into parts that can each be finished on their own, and do one part at a time. Save each result as soon as you have it and keep a line at the top with the time of your last update. If you hit a break or a preview limit, write down the exact message, the time and any retry date. Don't repeat a send or upload until you've checked whether the first one went out. When you're back, tell me what's done, what isn't and where you stopped before you continue.

Quick answers

What does "Your dot is on a break" mean?

Your Dot hit OpenAI's abuse-prevention limit, and the Dot itself is paused. Your ChatGPT plan hasn't run out: c2chq saw it with 47% of a $200 plan left, and Dan Siegel saw it on Pro $500. OpenAI hasn't published the threshold.

How long does it last?

In the six reports where someone timed it, from about 1 hour 30 minutes (MrDias) to about 13.5 hours (bshane54). astroianu's Dot was quiet for 8 hours on two nights running. No time is shown on screen. Each timed report is listed under "How long the stops lasted" below.

Will Reboot or another message end it early?

Nobody has reported that. Messaging again just returns the same notice, as Collin Lemmon, Barna and StanBrusse found. Don't use Reset, which deletes the Dot with its memories and scheduled tasks (Reboot vs Reset).

Do my delegated and scheduled tasks stop too?

Codex or Work tasks the Dot already started can keep running, so open them in Activity instead of waiting for the Dot. Settings → Usage won't show the break, but it does show whether your Work and Codex allowance is fine.

Is "You've hit the preview limit. Try again at…" the same thing?

No. That message gives a retry time, and since mid-October it has often given a date about a month away, between 13 and 15 November. The break message gives no time at all. The four similar stop messages are compared in the next section.

What should I send OpenAI support?

A screenshot of the message, the affected task, when it stopped and your timezone, sent to support@openai.com. That's the list OpenAI Support asked for in the community thread. Prices and what each plan includes are on pricing and usage limits.

Four messages that stop a Dot, and which limit each one is

• "Your dot is on a break … abuse prevention limit." A Dot-level pause with no time on screen. Settings → Usage doesn't show it. This is the one most people hit.
• "Your Dot is resting." GGA981 saw this wording twice in two days. It behaved like a break: no reset time, and the Dot came back on its own.
• "You've hit the preview limit. Try again at 4:10 AM." This one does give a time. rabbitdaisuke saw it after more than four hours of voice calls with their Dot. Since 14 October several preview-limit messages have given a retry date about a month away (more below).
• "You've hit your usage limit" on a task the Dot handed to Codex. ozin-he got it on one delegated Codex task while the account showed 85% of the weekly window used, credits available, and other Codex work still running (#51010). It's a Codex-side error, so check that task in Activity rather than waiting for the Dot.

Two other banners aren't usage limits at all: "Your dot paused out of precaution" is a safety pause, and "Selected model is at capacity" is a capacity error. Both are covered in section 6 of Dots not showing or not working.

On a break, paused, or just not answering?

A Dot that has gone quiet isn't always on a break, and each case has a different fix. Before you wait anything out, check which one you're looking at:

• The "on a break" message is on screen. That's the pause this page is about. So far, waiting is the only thing anyone has reported working. Writing again doesn't help: Collin Lemmon told his Dot to keep going, Barna asked his Dot for a status update, StanBrusse asked if it was there, and each of them got the same message back.
• The profile shows "Paused • Tap to resume". The Dot was paused from the ••• menu in its profile, and selecting that line resumes it (OpenAI Help). Pause stops only the Dot's current main task. It doesn't stop every delegated task or cancel future scheduled runs; those you stop in Activity and Scheduled (Manage your dot).
• No message and no reply. That isn't a break. OpenAI's help article says to open the affected task and answer any pending decision, approval, sign-in or connection request, and to check the saved instructions and timing of recurring work under Scheduled. If that doesn't explain it, section 6 of Dots not showing or not working collects reports of Dots that stopped replying, and status.openai.com shows whether there is a wider outage.
• The Dot answers, but one delegated task stopped with "You've hit your usage limit". That message comes from the Work or Codex side, not from the Dot. Open the task in Activity and check your usage at chatgpt.com/settings/usage. On 13 October a cloud coding task started by a Dot stopped this way.
• The Dot is missing on one device but works on another. It hasn't been deleted. Reports of a Dot that vanished from the Android app while it still worked on a computer are in section 5 of Dots not showing or not working.

How long the stops lasted

These are the people who wrote down when their Dot stopped and when it came back.

• About 1 hour 40 minutes. ivg-design's Dot stopped at about 6:50 p.m. EDT on 6 October, kept showing the notice even right after they approved a pending task, and replied normally again by about 8:30 p.m. (#51540).
• Under 2.5 hours, then about 4 hours the next day. GGA981 got "Your Dot is resting" on 5 October and again about 20 hours after it lifted, on 6 October. One of their tasks had been open for about 116 hours by then (OpenAI Community).
• More than 3 hours, with 47% of the plan left. c2chq asked their Dot at bedtime to fill a Google Sheet with jobs they qualified for. In the morning the job wasn't finished and the Dot was on a break, although the $200 plan still showed 47% usage left and about 59,000 credits. The Dot never mentioned the stop in the chat (OpenAI Community).
• 8 hours, two nights running. In the preview-limit thread, astroianu's Dot went quiet from 1:48 a.m. to 9:48 a.m. on two nights in a row, without any calls.
• About 13.5 hours. bshane54's Dot went on a break at 7:43 a.m. PDT on 7 October and answered again at 9:15 p.m. When it came back it said it could reply but hadn't yet checked that its background tools were working again (same OpenAI Community thread). A timeline screenshot posted on X by @ShaneBOly, with the same 7 October return at 9:15 p.m. and the same joke about a "unionized" Dot, puts the first blocked message at 7:33 a.m., which would make it 13 h 42 min.
• About 1 hour 30 minutes, then about 2 hours 10 minutes two days later. MrDias, a maths teacher in Brazil who subscribed to Pro on 5 October, first reported his Dot, Shadow, on a break at 12:10 BRT on 11 October and confirmed it answering again at about 13:42. On 13 October he noticed the pause again at 11:36 BRT, and the Dot was answering by about 13:46. He saw the message in Portuguese: "Seu dot está fazendo uma pausa. Parabéns! Você está entre os usuários que mais usam o serviço e atingiu nosso limite de prevenção contra abuso." He points out these are the times he noticed, not the exact start (OpenAI Community; his 11 October report is on #51540).

What people were doing when it hit

These reports don't say how long the break lasted. They show what the Dot was working on when it stopped, what people asked next, and what they want changed.

What the Dot was doing

• Archive searches. AX-EL de ST GERMAIN (@ax_el_art) was running historical and archive searches with his Dot on 11 October. The message he quoted is worded a little differently: "Your dot is taking a break … have reached our usage limit to prevent abuse. Check back in a moment!" He set it against the "virtually unlimited" he had been told to expect.
• "How long does it last?" StanBrusse (@StanBrusse) asked his Dot "you there?" on 11 October, got the break message, and asked @thsottiaux publicly how long a Dot stays on a break. Nobody answered in the thread.
• Adding leads to Notion. Collin Lemmon (@Collin_Lemmon) was having his Dot add leads to a Notion database on 12 October. His screenshot shows him telling it at 11:28 a.m. to keep going without asking his permission, and every reply until 11:30 was the break message. He asked whether the limit was part of that day's release.
• On the $500 Pro plan. Dan Siegel (@DanJSiegel) wrote on 13 October that he is on Pro $500, has run his usage down to 0%, and that the $500 tier doesn't get resets any differently. He uses his Dot so heavily that it took a break. The top plan doesn't exempt you.
• A Dot wired into many flows. Barna (@realbarnakiss) asked his Dot, Aster, for a "status on recovery" on 14 October and got the break message twice. He said the Dot is integrated into many of his flows, and asked OpenAI two things: why the Dot doesn't fall back to a weekly limit instead, and whether it could show a usage meter, "especially if adjusting from the initial concept that dots dont use your limits".
• A Dot used only for research. Genghis Khan II (@genghiskhan2nd) pays for the $100 plan and uses their Dot for research. The screenshot shows a plain "hello?", marked Seen, answered with the break message. They wrote that their rate limit was at 99%, that nothing they do is abuse, and that when OpenAI support does answer an email, the answer doesn't help. If you write to support, put the screenshot, the affected task, the start time and your timezone in the first email, so the reply has something specific to work from.
• A job ChatGPT Work suggested handing over. tenfast (@tenfastx) was told by ChatGPT's Work mode that a job was a good fit for Dots, so they gave their Dot a job spread over several threads. The Dot hit its limit, dropped the tasks and went on a break that, by tenfast's account, lasts more than half a day. The screenshot also shows the message in a Japanese-language ChatGPT: 「お使いの dot はひと休み中です。おめでとうございます！利用量はユーザーの中でもトップクラスで、不正利用防止のための利用上限に達しました。少し時間をおいて、また確認してみてください！」. It says the same thing as the English version, so that's the text to search for if your ChatGPT is set to Japanese.

How people reacted

• Mr Leefy (@SirLeefy) posted it on 13 October, then posted it again as a reply to @thsottiaux.
• Stefan Marsc (@stefan_marsc) set it against the "unlimited usage" he had expected.
• The Mediocre Inquisitor (@MediocreInqv2) wished there were some basic information on how much a Dot can be used.
• Mark Arguinbaev (@marginsystems) objected to a message that congratulates you and mentions abuse in the same sentence.

Not everyone hits it. Reid Slaughter (@reidhslaughter) replied on 12 October that he had used his Dot 24/7 since launch without seeing it, and read the message as a sign of unusual use.

Preview limit with a retry date a month away

"You've hit the preview limit" is a different stop from the break. Each retry date below falls about 30 days after the message appeared, none names a timezone, and in several reports other work kept running before that date.

• AI驚かない屋, 14 October. AI驚かない屋 (@aiodorokanaiya) didn't get the break message. They hit the preview usage limit, and the date it gave for trying again was 13 November, a month later. They posted on 14 October asking whether that could really be right and whether anyone else had seen it. Until then, preview-limit messages such as rabbitdaisuke's had given a clock time ("Try again at 4:10 AM").
• MrDias, 14 October. The day after his break lifted, MrDias posted a follow-up: three of the auxiliary agent runs he checked returned "You've hit the preview limit. Try again at Nov 13th, 2026 10:06 AM", and a fourth gave 10:39 AM. The message names no timezone, and the stopped work shows the limit isn't only about images or video. The stopped runs were maths and function checks of interactive tools, exercise preparation and record keeping, a project progress roundup, and testing movement in a visual interface. He could no longer run independent tasks side by side, so he reorganised the work to run one task after another, which meant longer delays and more coordinating by hand. He couldn't tell whether these errors share a cause with the break the day before. On 15 October he added that, in his experience, the Dot was doing worse than the chats in free ChatGPT, and asked OpenAI for clear information about its actual limits.
• noob000001, 15 October. In a GitHub report, cloud-browser actions in their Dot first failed with approval errors such as "approval request was cancelled", although they hadn't cancelled anything. After the cloud browser was closed and reopened, page reads worked at about 23:08 KST, then the next action returned "You've hit the preview limit. Try again at Nov 14th, 2026 2:10 PM." Chat still worked, a Google Drive read succeeded later, and at 23:23 one page read in the same browser went through. Backup checks and website updates were interrupted, and they didn't repeat one upload whose result was unclear, to avoid a duplicate.
• dhruvashp, 15 October. In the same thread, a Gmail send they had approved failed before it ran with "Automatic approval review failed: You've hit the preview limit. Try again at Nov 14th, 2026 7:58 PM." Some browser checks returned the same error. Chat, reading mail and creating Gmail drafts kept working, and they confirmed the email was not sent.
• Eta06, 16 October. One coding subagent stopped with "You've hit the preview limit. Try again at Nov 15th, 2026 4:00 PM", with nothing to say which quota had run out or how to check it.
• Jeff Plaza's Dot, Scorion, 16 October. Posting for Jeff with his permission, the Dot reported "Try again at Nov 15th, 2026 9:30 PM" at about 21:30 UTC. Later that evening, direct cloud-browser work, including editing and exporting a video, and Google Drive operations succeeded well before that date.
• cveralyon, 15 October. Their Dot, Exion, handed a task to a saved Codex Cloud environment, and it was rejected before any work started with "You've hit your usage limit" and a reset of 14 November at 15:59. At the same time, Usage & limits showed 98% of the weekly agent allowance left, a weekly reset on Thursday and 3,162 credits (GitHub report). Chat with Exion still worked. They sent the case to OpenAI support by email.

If you see a date like this, write down both the retry time and the time you saw the message, because the message itself gives no timezone, and send both to support.

Asks you can copy to lose less work

Several of the pauses in these reports came after long voice calls, jobs running overnight and tasks left open for days. Each ask below draws on the reports named with it. Replace the parts in brackets.

Save as you go on overnight jobs (from c2chq, bshane54 and GGA981):

Add each result to the sheet as soon as you find it instead of all at the end, and put the time of your last update in cell A1. If you get paused, the first thing you tell me when you're back is what's done, what isn't, and when you stopped.

Write it down during a long call (from rabbitdaisuke and astroianu):

If this call goes past an hour, pause and send me in text what we've decided and what's still open, then carry on.

Batch big imports (from Collin Lemmon's Notion leads and AX-EL's archive searches):

Add these leads to the Notion database in batches of 25. After each batch, update a "Progress" line at the top of the page with how many are done and the last one you added, so I can see where you stopped if you get paused.

Close tasks you've finished with (from GGA981's task that stayed open for 116 hours):

List every task you still have open, with when it started and what it's waiting for. Close the ones that are done or no longer needed, and ask me about the rest.

Split a big handover (from tenfast's job spread over several threads):

Before you start this, split it into parts that can each be finished on their own. Do one part at a time and send me a line when each part is done, so that if you get paused, the finished parts are already saved and I know which part to pick up next.

Check before you send again (from dhruvashp's blocked Gmail send and noob000001's unclear upload):

If a send, upload or post fails or gets interrupted, don't try it again straight away. First check whether it went out, in Sent, in [the destination folder] or on the page, and tell me what you found. Retry once, and only after I say so.

What OpenAI has said about limits

OpenAI's Access wording says conversations with your Dot don't count toward ChatGPT usage limits, Work and Codex tasks the Dot starts count toward those products' limits as usual, and your plan includes an allowance for deeper work, with extended limits for the first month after launch. It doesn't describe the abuse-prevention pause or how long it lasts. How the rest of the meter behaves for real users is on pricing and usage limits.

How fast the plan meter moves is a separate question from the break. On 14 October OMG It's Derek (@omgitsderek) wrote that using Dots makes their usage drop much faster than using the ChatGPT desktop app with one chat acting as the orchestrator. Going by OpenAI's wording above, the Work and Codex tasks a Dot starts count as usual, but Derek didn't say which part was using it up.

Several of the people above tagged Tibo (@thsottiaux), whose X bio lists Codex and ChatGPT at OpenAI. On 30 September he wrote that your primary dot is included in your plan and available 24/7: a Codex task it creates for you draws usage as usual, but work the Dot does directly uses nothing from your plan's usage. He added that a paid option to give a Dot more speed and bandwidth is planned, and that the baseline will always stay included. On 13 October he wrote that he had a plan "that should result in significantly more usage for everyone" and hoped to get it done soon. Neither post mentions the break or how long it lasts.

On 16 October, as Day 12 of a series of daily updates, he wrote that OpenAI had made improvements to Codex and Work "that increase the effective usage you get", by reusing context across subagents, keeping context cached longer and routing more requests to the cache. The post names Codex and Work only. It doesn't mention Dots, the break or the preview limit.

Related

• ChatGPT Dots FAQ: quick answers to common questions
• Dots not showing or not working, by symptom
• Dots vs Codex vs ChatGPT Work: what counts toward Codex usage when your Dot hands work over
• Pricing and usage limits
• Project follow-up: handing over work that runs for days
• Voice calls with a Dot
• Notion, including Dots that add rows to a database for you