How do you fix ChatGPT's too many requests error?
Run these six steps in order. The cheap ones come first, and none of them deletes a chat.
- Stop clicking for five minutes. The message asks for "a few minutes", and every reload or sidebar scroll loads conversations again.
- Close every extra ChatGPT tab and window, including the desktop app if it's open next to the browser. Keep one.
- Pause any chat cleanup. Archiving, deleting, renaming and scrolling far down the sidebar all load conversations, and forum posters hit the lock doing exactly that.
- Turn off extensions that export, sync or organize your chats, and stop any script that drives ChatGPT for you.
- Turn off your VPN or proxy. OpenAI's help page warns that a VPN can route your traffic "through flagged IPs".
- Check status.openai.com, then contact support if the lock follows you across devices for hours. Send the exact message, the times with your time zone, and a HAR file.
Here's the full 2026 wording, quoted in an OpenAI community bug report on 3 October 2026: "Too many requests. You're making requests too quickly. We've temporarily limited access to your conversations to protect your data. Please wait a few minutes before trying again." Read the middle sentence again. ChatGPT has limited access to your conversations. It names no model and no reset time. A plan allowance shows its reset time when ChatGPT has one, per OpenAI's model help page.
We searched the OpenAI developer community on 11 October 2026 and kept every thread from June to October 2026 whose title or first post was about this error or its HTTP 429 code. That left 9. No poster settled on fast prompting as the cause. One user only scrolled the sidebar. Another first suspected their own plugin tests, then got the lock on a page where they hadn't sent a prompt at all. Two feature requests from July and August 2026 aren't in the 9, because they ask for multi-select delete and archive. Both say deleting or archiving chats one by one tripped the same lock.
Which too many requests message are you seeing?
The wording tells you which problem you have. This table covers the ChatGPT apps only, as of October 2026, with where each string was reported. API errors have their own table further down.
| Message on screen | Where it was reported | What it points to | First fix |
|---|---|---|---|
| "Too many requests. You're making requests too quickly. We've temporarily limited access to your conversations to protect your data. Please wait a few minutes before trying again." | Web, Windows and Mac apps, Android; OpenAI forum, June to October 2026 | A short lock on loading your chats after a burst of activity | Steps 1 to 3 above |
| "Too many requests" alone, in a small notice | Windows desktop app, August 2026; a forum reply on 11 October 2026 | The same lock in a shorter banner | Wait, then reload once |
| "Too many requests in 1 hour. Try again later." | OpenAI forum, December 2022, during the free research preview | An hourly cap from 2022; none of the 2026 threads quote it | Treat it like the lock above |
| "We're currently processing too many requests. Please try again later." | OpenAI forum, February 2023 | A refusal on one account while the poster's friends could still use ChatGPT | Check the status page, then wait |
| "Too many concurrent requests" | IT Daily, 23 October 2025, during a ChatGPT outage | More than one reply running at once, or an OpenAI incident | Let the running reply finish; ChatGPT limit reached has the concurrency fixes |
| "We detect suspicious activity." or "Unusual Activity Detected" | OpenAI help center | Traffic that looks automated, often a flagged VPN address | VPN off, private window, stop sharing the account |
| A limit message that names a model or a reset time | OpenAI help center | A plan allowance on one model or tool | ChatGPT limit reached |
The 2022 string is the one most older guides still quote. A December 2022 forum thread has a user who got it after "only two questions" in a day. OpenAI never published how long that hour lasted. Current plans print no hourly cap for text: OpenAI's pricing page lists everyday text chats as "Unlimited" on Free, Go, Plus and Pro.
If the screen says "Something went wrong" instead, that's a different error with its own fix list in ChatGPT something went wrong.
What triggers the too many requests error in ChatGPT?
OpenAI's abuse safeguards trigger it, and OpenAI says so in its model and limits help page, updated in October 2026: "OpenAI uses safeguards to help prevent misuse, which may occasionally result in a temporary usage restriction. ChatGPT notifies you if this happens." The pricing page says it on the Free, Go and Pro cards too. Text chat is unlimited, "subject to abuse guardrails". OpenAI doesn't publish what the guardrail counts. The forum threads show what tripped it in practice.
Chat cleanup. In June 2026 one user got the lock every five archive or delete actions, and finished the job in the mobile app. Another hit it just by scrolling the sidebar while deleting and renaming old chats. It kept coming back for four days, until the cleanup was done.
Several tabs at once. A Pro subscriber who works in two or three tabs at a time got locked on 7 August 2026, often right after an image finished. OpenAI support replied in that thread on 10 September and offered to "pass along the feedback on these controls". A September post asked OpenAI for a "Trusted High-Activity Mode" for people who work across many chats on purpose. OpenAI hasn't replied to it.
Extensions and scripts. OpenAI's Terms of Use, effective 1 January 2026, ban tools that "automatically or programmatically extract data or Output" and any attempt to "circumvent any rate limits". An exporter that opens every chat in your history makes the same requests as scrolling the sidebar, only faster. Turn it off before you blame anything else.
VPNs and shared addresses. OpenAI ties flagged IP addresses to the suspicious activity message, not to this one. Still, a VPN exit shared with thousands of strangers is a cheap thing to rule out. Turn it off for ten minutes and see.
OpenAI incidents. After a ChatGPT incident on 3 September 2026, a Plus user's large Project kept failing with repeated HTTP 429 responses. On 2 October OpenAI support told a Business user with the same errors: "These errors may be linked to a recent outage." If the status page shows an open incident, nothing on your side will help.
How long does ChatGPT's too many requests block last?
ChatGPT's block lasts "a few minutes", in the message's own words, and that matches most reports. It runs longer when you keep doing the thing that tripped it. Here's what posters reported, each in their own 2026 thread.
| Thread opened | What the poster was doing | How long it lasted |
|---|---|---|
| 1 June 2026 | Unarchiving and deleting archived chats | Came back every 5 actions; finished in the mobile app |
| 15 June 2026 | Scrolling the sidebar, deleting and renaming chats | Gone after 16 hours, then back; cleared 19 June when the cleanup ended |
| 5 August 2026 | Normal use in the Windows desktop app | Reloading or waiting a few minutes cleared it; it kept recurring |
| 7 August 2026 | Image work in two or three tabs, Pro plan | Stopped 3 to 4 hours later; back on 10 August |
| 9 September 2026 | One tab: uploading a screenshot, copying the reply | Cleared after a few minutes, came back on the next copy |
| 3 October 2026 | Opening the custom GPT migration page, no prompt sent | Repeated through the day |
So wait five minutes, then change what you were doing. A lock that survives one tab, no extensions, no VPN and a different device for hours isn't yours to fix. That's when OpenAI support wants the exact message, the timestamps and a HAR file.
Does paying for ChatGPT Plus or Pro stop too many requests?
No. Four of the 9 threads came from paying accounts: two on Plus, one on Pro and one Business seat. A fifth poster upgraded mid thread and wrote that it seemed "to be worse now". As of October 2026 ChatGPT costs $0 on Free, $8 on Go, $20 on Plus and from $100 on Pro. The pricing page marks everyday text chats "Unlimited" on all four, with a footnote that usage "must be reasonable".
Whizi, which publishes this page, runs the same kind of guard and prints the numbers: 10 chat messages a minute and 60 an hour per user, the same on every plan. Its per-minute error reads "Too many requests. Please wait and try again." and the hourly one reads "Hourly message limit reached. Please try again later." A refused request spends none of your allowance (Whizi rate limits). Where Whizi loses: that 60 an hour is a hard ceiling, and ChatGPT prints no hourly number for text chat at all.
How do you fix a 429 too many requests error from the OpenAI API?
Read the error code before you retry, because the OpenAI API sends 429 for three different problems. OpenAI's API 429 help page says it plainly: "A 429 response can indicate a temporary rate limit, an exhausted prepaid balance, or a spending or usage limit." Only the first one clears with time.
| Error or code (OpenAI API, October 2026) | What it means | Does a retry help? | Fix |
|---|---|---|---|
| Rate limit reached for requests or tokens | Your organization or project passed its requests or tokens per minute | Yes, after a delay | Wait at least the Retry-After header, then back off with jitter |
slow_down (type rate_limit_error) | Your traffic rose too fast, even under your per minute limits | Yes, slowly | Follow Retry-After, then ramp up gradually |
credit_balance_exhausted | No prepaid credits left | No | Add credits in API billing |
organization_usage_limit_exceeded | You reached the usage limit OpenAI assigned | No | Request a higher limit |
organization_spend_limit_exceeded or project_spend_limit_exceeded | You reached a spend cap you set | No | Raise or remove it, or wait for the monthly reset |
Two details catch people out. A limit of 60 requests per minute "may also be enforced over 1-second periods", so a burst of five calls can fail while your average looks fine. And failed calls count too. OpenAI warns that "unsuccessful requests contribute to per-minute limits", so a tight retry loop makes it worse.
Limits belong to the organization and project, not to you as a user. They rise with your total credit purchases, per OpenAI's rate limit guide: Build at $5, Launch at $100 and Grow at $500, as of October 2026. Your current numbers are on the Limits page of your organization settings.
- Wait five minutes without reloading, because the message asks for a few minutes and every reload loads conversations again.
- Keep one ChatGPT tab or app open at a time while the lock is on.
- Archive, delete and rename old chats in small batches, or finish the cleanup in the mobile app.
- Turn off chat exporter and organizer extensions, and any VPN or proxy.
- Check status.openai.com before you change anything else.
- On the API, read error.code first: only a rate limit clears with time, and billing or spend errors never do.
Frequently asked questions
Why does ChatGPT say too many requests when I barely used it?
Because the lock reacts to how fast your chats load, not to how many prompts you send. Forum posters in 2026 tripped it by scrolling the sidebar, archiving chats, copying one reply and opening a page with no prompt sent. Close extra tabs, pause the cleanup and wait five minutes.
How long should I wait after "Too many requests in 1 hour. Try again later."?
Start with five minutes. That hourly message comes from ChatGPT's 2022 research preview, OpenAI never published its length, and the 2026 wording asks for "a few minutes". If it comes back after that, find the trigger (extra tabs, chat cleanup, an extension) rather than waiting longer.
Is too many requests the same as hitting my ChatGPT usage limit?
No. A usage limit is a plan allowance on one model or tool, and OpenAI says ChatGPT shows when it resets if that information is available. Too many requests is a short anti-abuse lock that names no model and no reset time. The ChatGPT limit reached guide covers the allowance messages.
Can a VPN cause the ChatGPT too many requests error?
OpenAI doesn't list a VPN as a cause of this message. It does for the neighboring one: its help page says VPNs and proxies can route traffic "through flagged IPs", which triggers "We detect suspicious activity." Turning the VPN off for ten minutes costs nothing and rules it out.
