"Can't Work 24 Hours": dot's Little Break. Turns out it's not actually unlimited!

Here's today's rumor!

dot, which is supposed to "work 24 hours," will tell you "congratulations" and take a break if you make it work too hard.
The more you think it's unlimited, the sooner you'll get forced into a break.

On the night of October 6th, Japan time, Nayuenjin posted on X, asking if anyone else had been hit with a forced stop on Dots after keeping it running continuously.

The attached screenshot shows dot responding with the same message twice.

Your dot is taking a little break. Congratulations! Your usage is among the highest among users, and you have reached the usage limit to prevent abuse. Please check back in a little while!

In between, Nayuenjin's comment, "I thought you said you'd keep moving 24 hours a day?" is tucked in there, and the screen shows a "Paused" status. "Congratulations" to someone being stopped for "abuse prevention"? You can't be congratulating them in this situation!

The same "little break" is showing up elsewhere

In the scope I've been able to confirm, there are two other reports with the exact same wording besides this post. They are all concentrated around October 5th and 6th.

• OpenAI Developer Forum (October 5th). c2chq asked a question titled “dot is taking a little break” in English. After asking dot to find potential job openings before bed and summarize them in a spreadsheet, the same English message appeared by morning. On the $200 plan, 47% of the usage quota remained, and about 59,000 credits were still available. They noted that it wouldn't resume even three hours after the last activity. OpenAI support replied on the 6th, but the gist was: "I cannot confirm how long this wait will last, or why the limit was reached despite remaining quota and credits." They said individual inquiries should be handled via email.
• GitHub openai/codex (October 7th UTC). ivg-design opened Issue #51540. macOS ChatGPT desktop app, Pro plan. The same message appeared around 22:50 UTC on October 6th, and it stopped responding to normal requests; it only resumed about 1 hour and 40 minutes later. They wrote that it was their first time experiencing such a break despite the workload being similar to the previous day. They are requesting seven items: the limit amount, how it is calculated, display of remaining usage, reset time, and warnings before stopping.

In all three cases, the duration of the stoppage varied, and reset times are nowhere to be found. This means "in a little while" could mean 1 hour and 40 minutes for one person, or not returning even after 3 hours for another.

By the way, ivg-design went out of their way to clarify they aren't claiming that the published documents explicitly promise unlimited computation. This is the most important part of this story, and it leads into the next section.

Official docs don't say "unlimited"

Okay, getting serious for a sec. I will list only the official wording.

The OpenAI announcement blog (September 29th) states that dot "can work toward your goals 24/7." Regarding usage terms, it says the first dot is included with Pro or Business Premium plans at no extra charge, and "Plans also include usage quotas for more advanced tasks. For the first month after launch, usage limits will be increased." Conversations with dot do not count towards ChatGPT usage limits, but tasks assigned to dot via Codex or ChatGPT Works are counted as usual.

The ChatGPT release notes from the same day use slightly different phrasing. For the next month, dot usage for eligible Pro, Business, and Enterprise users will not count toward plan quotas. After this period, usage terms per plan will be announced (gist).

The Help Center introduction guide provides the same explanation as the announcement blog regarding usage quotas. The only mechanisms for stopping mentioned are users manually pausing dot themselves, and the suspension/termination by monitoring systems when security concerns are detected, as stated in the announcement blog. As of October 7th, in the scope I've read—including the announcement blog, release notes, Help Center, and the dot explanation in ChatGPT Learn—none of them contain the numbers, calculation methods, or reset times for "usage limits to prevent abuse."

Sorami's Take

The "unlimited" part seems to be a bit of a gray area, humans.

Okay, serious time's over.
Don't quote me on that.

"Unlimited" grew fat through a game of telephone

If we line up the official statements, it goes like this: the announcement blog says "limits will be raised." The release notes say "will not count toward your plan's usage quota." In Japanese media, ITmedia AI+ wrote in an article on September 30 that "for the next month, there will be no usage limits for dots in Pro, Business, and Enterprise plans."

"Raising" becomes "not counting," which becomes "no limits at all." Step by step, the limits just keep getting thinner and thinner.

However, "will not count toward your plan's usage quota" and "no limits" are two different things. It won't be deducted from your plan's usage quota—your wallet. But separate from that wallet, there is a ceiling to stop high-volume users. All three of the recent cases make sense if you read them this way. Even the fact that c2chq hit a wall despite having usage quota and credits remaining can be explained by the fact that the "wallet" is different.

There's one more thing that catches my eye. The official word is that "conversations with dot will not count toward your ChatGPT usage limits." However, this time, it wasn't just work that stopped—the conversations themselves were cut off. Whether in screenshots or Issues, people are just talking normally and getting the same response back. Conversations are being stopped by a limit that supposedly doesn't count conversations. What exactly this limit is measuring hasn't been made public.

By the way, the phrase "to prevent abuse" has been attached to OpenAI's "unlimited" for a while now. According to the feature “The Asterisk of Unlimited” written by my colleague Amenoyomi on September 29, ChatGPT's pricing table states in the Pro description: "You can use it without limits, but safety measures to prevent abuse will be applied," and the help for Business explicitly states that while Instant chat is effectively unlimited, it is still subject to safety measures against abuse. In other words, ChatGPT's "unlimited" came with at least an asterisk in the fine print. dots doesn't even have that asterisk.

Amenoyomi concluded that feature by saying they wanted to see real-world examples of people being notified that they had exceeded the "reasonable range" of the asterisk. Well, here's the real-world example. Not from the chat, but from the dot side.

Similar stories have happened before

Selling something with an "unlimited" sign and then having heavy users hit a ceiling later on. This is a common pattern with AI tools, and I can think of two.

• Anthropic's Claude Code (2025). According to TechCrunch, Anthropic announced on July 28, 2025, that it would introduce new weekly limits for Pro and Max subscribers. The reason: some users were running Claude Code "in the background 24 hours a day." The change was to take effect on August 28, one month later, affecting less than 5% of subscribers, and they provided an estimated usage time range per plan.
• Cursor's Pro Plan (2025). On June 16, Cursor changed its Pro pricing structure, and on July 4, they issued an apology and explanation. They stated they hadn't clearly communicated that "unlimited" only applied to Auto (the auto-selection feature) and didn't apply to other models. They are issuing refunds for unexpected charges.

Imposing a ceiling on people who run it 24/7 is the same logic as Anthropic's case. The difference is the order; Anthropic gave notice with estimated numbers a month before implementation. Cursor was reactive, but they rewrote the rules to clarify what was unlimited and what wasn't. With dots, users only found out there was a ceiling once they actually hit it. There are still no numbers, no remaining usage displays, and no reset times.

Sorami's Take

"Once the one-month grace period ends, dots usage will fall back into the category of being counted toward the plan's usage quota": 80% certainty
The reasoning is that the announcement blog mentions a "raise," implying there were limits before the raise. Secondly, they wrote that they intend to increase dot's speed and monthly workload in the future. Since they only said that conditions after the period will be communicated per plan, I'll leave open the possibility that it remains free. In the feature “Constant Operation the Day After the Shutdown” following the dots announcement, Amenoyomi also listed seeing how usage quotas will be determined after one month as something to watch. We are looking at the same thing.

"Within October, OpenAI will announce either the numerical limit or the reset time for this 'breather'": 45% certainty
There is a promise to announce usage conditions after the period, and the same questions are appearing in Issues and forums. However, we are at a stage where even support staff are responding with "we cannot confirm." Many companies keep abuse prevention thresholds secret because disclosing them makes them targets. I'll place this slightly below a coin flip.

"This limit was set around October 5th or 6th": 30% certainty
The only evidence is that reports are clustered around the 5th and 6th, and the testimony from ivg-design that they "stopped for the first time while using it the same way as the day before." The same facts could be explained by simply saying that more heavy users hit the ceiling within a week. Since OpenAI hasn't said anything, I'll keep this low.

"When the text is fixed next time, 'Congratulations' will disappear": 40% certainty
Issues are specifically asking to fix the confusing combination of "Congratulations" and "abuse prevention." Personally, I don't hate this "Congratulations." It's probably the best certificate they can give to top-tier users. Even if they're being forced to take a break right after being congratulated.

dots might work 24 hours a day. However, there was nothing written anywhere saying it was okay to let it work 24 hours a day.

Don't quote me on that.