“Too many concurrent requests” and “Reached concurrency limit” both mean that ChatGPT is refusing new work because it counts too much activity in progress at the same moment. “Concurrent” simply means “at the same time.” That count can be too high for two very different reasons. Either OpenAI's own systems are struggling, and nothing you do on your device will help, or something tied to your account is still running or stuck.
That is why the order of your checks matters more than any single trick:
- Stop resending. Give the current reply a minute to finish or fail.
- Open status.openai.com. If there is an incident that matches your plan or region, wait for it to be resolved.
- If there is no incident, end everything that overlaps: other tabs, the desktop or phone app, uploads, long-running tasks, extensions.
- If nothing overlaps, sign out, clear ChatGPT's site data, and sign back in.
- If the message follows you across browsers, devices, and networks, contact OpenAI support.
One fact frames everything below. As of October 1, 2026, OpenAI's Help Center has no article for either message, and its guide to ChatGPT error messages lists neither wording. OpenAI also publishes no number of simultaneous requests, tabs, sessions, or conversations for any ChatGPT plan. A fixed figure such as “X requests per hour” or “two sessions on Plus” has no OpenAI source behind it.
What does “Too many concurrent requests” mean on ChatGPT?
It means the service believes more requests from you are in flight right now than it will accept, so it rejects the newest one. It is not a statement about how much you have used ChatGPT today. A usage cap counts messages over hours. A concurrency limit counts what is running this second. That is why the message can appear on your first prompt of the day, and why waiting for a daily or weekly cap to reset is the wrong remedy.
Because OpenAI does not document the message, it helps to keep apart what is documented, what has been reported, and what is only a reasonable guess.
| Statement | How well it is supported |
|---|---|
| No help article covers either wording, and no plan has a published concurrency number | OpenAI's Help Center, as of October 1, 2026 |
| The message was shown to large numbers of users during outages on June 10, 2025, October 26, 2025, and July 25, 2026 | Reported by El Mundo (June 2025) and by the incident log at did2memo (all three dates) |
| People get it with a single tab open, or after days of not using ChatGPT, and say that signing out and clearing site data fixed it | User reports in Reddit threads and how-to videos; no success rate is known |
| A second message sent mid-reply, several tabs or devices generating at once, uploads, long tasks, extensions, or a shared login can trigger it | Inferred from the wording of the message; not a rule OpenAI has stated |
| The exact threshold, how long the block lasts, and whether a paid plan raises it | Unknown |
OpenAI does confirm one related thing. Its help page on the separate Suspicious Activity Alert lists “Multiple Concurrent Sessions: More simultaneous logins than usual” as a trigger. So simultaneous sessions on one account are watched. That page does not say they cause the concurrency message.
Match the exact message on your screen to a first action
The wording on screen tells you where to start. The strings below are the ones people report seeing. OpenAI documents none of them, so read the right-hand column as the most sensible first move, not as an official instruction.
| Message on screen | What the wording tells you | Start here |
|---|---|---|
| “Too many concurrent requests” | No cause and no wait time given | Stop resending, then check the status page |
| “Reached concurrency limit” | The same condition in different words | Same as above |
| “Failed to load subscription: too many concurrent requests. Contact support@openai.com if error persists.” | It failed while loading your account details, not while answering a prompt, so your own prompting is an unlikely cause | Status page, then sign out and clear site data. The message itself points to support if it stays |
| “Unable to load GPT: too many concurrent requests” | It failed while opening a custom GPT | Reload once, test a normal chat, then check the status page |
| “Your current concurrent conversation limit has been reached. Please try again later.” | It names conversations, so several chats or tasks are probably running at once | Stop or finish the other conversations and tasks first |
| “You are making requests too fast! Please wait ~10 seconds...” | It states its own wait: about 10 seconds | Wait, then send once |
| “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.” | It states its own wait: a few minutes | Stop everything, extensions included, and do not keep reloading |
The subscription message also circulates with a different ending: “visit help.openai.com and open the chat bubble to contact support.” Treat it the same way.
How to fix ChatGPT too many concurrent requests: five checks in order
Work from the top and stop as soon as one check explains the error. Steps 1, 2, 4, and 5 use the general troubleshooting actions OpenAI lists for other ChatGPT errors. OpenAI does not say they cure this particular message, and none of them comes with a tested success rate.
1. Stop resending and let the current reply finish
Every extra click on Send or Regenerate is one more request in flight. If a reply looks stuck, OpenAI's advice for a hanging response is to wait 30–60 seconds, click “Stop generating,” and then regenerate once. Do exactly that, and no more.
Stop here if the next single message goes through. You had an overlap, and it has cleared.
2. Check status.openai.com for an incident that matches you
Open status.openai.com before touching your browser. Incident titles never quote the on-screen error. Look instead for wording such as “elevated errors” or “increased error rate,” and check which plan or region the title names.
This check deserves second place because incidents are frequent. OpenAI's status history lists at least 20 incidents on 15 different days in September 2026, counting titles that name ChatGPT errors. Some were scoped narrowly: “Elevated errors in ChatGPT conversations for Free and Go plans” on September 1, “Elevated errors for ChatGPT users in Europe” on September 11 and 14, and “Increased error rate for Plus and Pro users” on September 22. Dates on that page follow your local time zone and can shift by a day.
Stop here if an open incident matches your plan or region. Clearing cookies or reinstalling the app will not help. Wait for the incident to be marked resolved, then try once.
3. Clear overlapping activity on your account
With no incident showing, look for work that is still running under your login. These are the likely sources of overlap:
- a new message sent while the previous answer is still streaming
- several tabs, the desktop app, and the phone app generating at the same time
- a file that is still uploading or being processed
- a long-running task such as an agent run or Deep Research
- a browser extension or automation tool that calls ChatGPT in the background
- another person using the same login
Close the extra tabs and apps, let long tasks finish or cancel them, and wait for uploads to complete. If someone else may be signed in, use OpenAI's “Log out of all devices” option, then sign in on one device.
Stop here if ChatGPT works once only a single conversation is active.
4. Sign out and clear ChatGPT's site data
If you had one tab open and nothing else running, a stale session in the browser is the next suspect. This is the fix users most often report.
A private or incognito window makes a quick test. If ChatGPT works there, the problem is in your regular browser profile, either in its stored data or in an extension. Then do the full reset:
- Sign out of ChatGPT.
- Delete the cookies and site data stored for chatgpt.com and openai.com. In Chrome, open the site information panel at the left of the address bar and go to Site settings. In Firefox, use “Clear cookies and site data” from the padlock menu. In Safari, go to Settings, then Privacy, then Manage Website Data. Menu names vary a little between browser versions.
- Close the browser completely, reopen it, and sign in again in a single tab.
If you were using ChatGPT without signing in, sign in and try again. During the June 2025 outage, the author of the did2memo log saw the message far more often on logged-out sessions. That is one observer on one day, so treat it as a cheap test, not a rule.
Stop here if the first prompt after signing back in gets an answer.
5. Change one thing at a time: browser, network, device
If the reset did not help, change a single variable and test after each change:
- Turn off browser extensions, especially privacy and security tools.
- Turn off any VPN, proxy, or secure DNS service.
- Switch networks, for example from Wi-Fi to a phone hotspot.
- Try a different device.
On a company network, OpenAI's network recommendations suggest two questions: is it only your machine or everyone on the network, and does the error disappear on a cellular hotspot? If the hotspot works, ask IT to confirm that *.chatgpt.com, *.openai.com, *.oaistatic.com, and *.oaiusercontent.com are allowed, along with WebSocket traffic to wss://ws.chatgpt.com on TCP 443.
Stop here if one change makes the error go away. You have found the cause.
Reached concurrency limit in the ChatGPT app or on iPhone
The same order applies in the desktop and mobile apps, with two differences. There is no site data to clear, so the equivalent of step 4 is to sign out of the app, close it fully, reopen it, and sign back in. The network test is also quicker on a phone: switch between Wi-Fi and cellular data and send one message.
It also helps to open ChatGPT in a browser on a different device. If the web version works and the app does not, the problem sits with that app or that device. If both fail with the same message, go back to the status page and to the overlap check in step 3.
How long do you have to wait?
OpenAI does not document a duration. Figures such as “five to ten seconds” or “one to five minutes” that circulate online have no source. What can be said depends on which situation you are in:
- The message names its own wait. Two variants do: about 10 seconds for “You are making requests too fast!”, and a few minutes for the “temporarily limited access” version. Wait that long, then send one message.
- You had overlapping work. The block should lift once that work finishes or is stopped. This follows from the wording of the message and is not a documented rule. Wait a minute or two after closing everything, then try once.
- OpenAI has an incident. The wait lasts as long as the incident. On June 10, 2025, El Mundo reported the message appearing from early morning, with some users back to normal at about 19:00 Spanish time.
If a couple of minutes of real quiet do not clear it and the status page is clean, move on to the next step. More waiting will not tell you anything new.
Does ChatGPT Plus or another paid plan stop the error?
Nothing OpenAI publishes says that Plus, Pro, or Business raises a concurrency threshold, so upgrading only to get rid of this message is a gamble. What a paid plan changes is usage allowances and model access, and those are different limits from the one this message describes.
Paid accounts are not shielded from the outage-driven version either. Several September 2026 incidents were scoped to paying users: “Increased Error Rate For Pro and Plus Plan Conversations” on September 10, “Elevated errors from GPT-5.6 and GPT-5.6 Instant on paid plans” on September 15, and “Elevated Error Rates for ChatGPT across Plus and Pro plans” on September 23.
If you are weighing an upgrade for other reasons, decide on those reasons. The guide to choosing between ChatGPT Plus, Pro, and Business covers who each plan is for.
When to stop troubleshooting and contact OpenAI support
Contact support when the message appears on more than one browser, more than one device, and more than one network, while the status page shows no incident. At that point the cause sits with your account, and no further change on your side is likely to matter.
OpenAI asks for specific material in that case. Collect these before you open the chat bubble at help.openai.com:
- a HAR file and the browser console errors, with timestamps
- the model you were using
- the conversation URL or ID
- the exact wording of the message, and what you were doing when it appeared
The “Failed to load subscription” variant names support in its own text. If it survives a sign-out and a site-data reset, skip the remaining steps and write to them.
API users: a 429 has more than one cause
If you call the OpenAI API and came here because of a 429, the rules are different, and they are documented. A ChatGPT subscription does not change API limits.
According to OpenAI's guide to API rate limits and 429 errors, a 429 “can indicate a temporary rate limit, an exhausted prepaid balance, or a spending or usage limit.” Read the error code before you retry. Codes such as credit_balance_exhausted, organization_usage_limit_exceeded, organization_spend_limit_exceeded, and project_spend_limit_exceeded are billing or budget problems, and no amount of retrying fixes them.
For a temporary rate limit:
- Honor the
Retry-Afterheader when it is present. Otherwise use exponential backoff with jitter, and cap the number of retries. The official SDKs already do this. - Do not resend immediately. Failed requests count toward your per-minute limit, so a tight retry loop prolongs the block.
- Smooth out bursts. Limits can be enforced over shorter windows than the one displayed: OpenAI's example is that 60 requests per minute may also be enforced over one-second periods. Sending many requests in parallel can therefore fail even when your per-minute total is fine.
- Remember that limits apply to the organization and the project, not to each user. A teammate's batch job can use up the capacity you were counting on.
Your actual limits vary by model and usage tier and appear on the Limits page of your organization's settings. If the 429 shows up inside Codex after its own retries, see Codex 429 After Retries: Find the Limit That Owns the Request.
Related questions about the ChatGPT concurrency error
Why does ChatGPT say too many concurrent requests when I only have one tab open?
A single tab does not rule out either cause. OpenAI may be having an incident, which is what happened on the three outage dates listed above. Your browser may also be holding a stale session, which is why users who hit the message after days of not using ChatGPT report that signing out and clearing site data fixed it. Check the status page first, then go to step 4.
Is “Too many concurrent requests” the same as a ChatGPT message limit?
No. A message limit counts how many messages you have sent on a model over a period, and it resets with time. The concurrency message is about requests in progress at the same moment, so it can appear when you have barely used ChatGPT. Waiting for a usage cap to reset does not address it.
Does the concurrency error mean my account was hacked or flagged?
Not by itself. OpenAI's Suspicious Activity Alert is a separate notice. If you share your login, or you see activity you do not recognize, follow OpenAI's steps for that alert: change your password, turn on two-factor authentication, log out of all devices, and sign back in on one trusted device and network.
What if my error looks similar but is not one of these?
Two neighboring problems have their own fixes. If a reply breaks off halfway with a stream error, see ChatGPT Error in Message Stream? Save Your Work, Check Status, and Fix the Right Branch. If your sidebar went empty and a “too many requests” notice appeared after you deleted a chat, see ChatGPT History Went Blank After Deleting a Chat: How to Triage Too Many Requests.
