ChatGPT error 429 means the service thinks you have sent too many requests in a short period, exceeded a usage limit, or triggered a rate-limit rule tied to your account, session, IP address, browser, workspace, or API organization. The fastest fix is to stop retrying for a few minutes, refresh once, sign out and back in, disable VPN or proxy routing, and check OpenAI’s status page before changing anything else. If you are using the OpenAI API, the correct fix is different: read the response message and headers, slow the request rate, reduce token demand, add exponential backoff, and raise limits only when the workload truly needs it.
What ChatGPT error 429 means
ChatGPT error 429 is a rate-limit error. In HTTP terms, 429 Too Many Requests means the client sent too many requests in a given amount of time, and the server is asking the client to slow down.[1] In OpenAI’s API documentation, a 429 can mean requests are arriving too quickly, or that the account has exceeded quota or billing-related usage limits.[5]
For a normal ChatGPT user, the visible message may not say “HTTP.” It may say “Too many requests,” “You are being rate limited,” “Try again later,” or a plan-specific usage-cap message. The practical meaning is the same: ChatGPT is refusing more work for now. Repeatedly pressing send usually makes the problem last longer.
For an API developer, a 429 is more precise. OpenAI says rate limits are measured across requests per minute, requests per day, tokens per minute, tokens per day, and images per minute, and the first exhausted limit can trigger the failure.[3] OpenAI also notes that limits vary by model and can apply at the organization and project level, not only to one user.[3]
The most important first step is to identify which product path you are on. If the error appears in the ChatGPT web or mobile app, treat it as a user-session or plan-limit problem first. If it appears in code, logs, an automation tool, a plugin, a custom GPT Action, or a server response, treat it as an API rate-limit problem first.
Quick fixes for ChatGPT users
Start with the least destructive fixes. A 429 is usually temporary, so avoid deleting chats, reinstalling the app, or changing account settings before you confirm the simple causes.
Stop sending requests for a short period
Close extra ChatGPT tabs and wait. OpenAI’s Help Center says continuously resending a failed request will not work because unsuccessful requests can still count against the per-minute limit.[2] If you have been rapidly regenerating answers, uploading files, asking follow-up questions, or switching tools, give the limit time to reset.
Refresh once, then sign out and back in
After waiting, refresh the page one time. If the error persists, sign out, close the browser tab, reopen ChatGPT, and sign in again. This helps when the browser has a stale session or a stuck request loop. If you cannot sign in at all, use our ChatGPT login failed guide instead because login problems often have a different cause.
Check whether ChatGPT is having a wider incident
OpenAI’s status page is the best first-party place to check whether ChatGPT or API components are degraded, although it also warns that individual availability can vary by subscription tier, model, and API feature.[6] If the status page shows an incident, waiting is usually better than changing your local setup. For a broader view of recent reliability events, see our chatgpt outages 2026 timeline and ChatGPT Outage History.
Disable VPN, proxy, or shared network routing
Rate-limit systems can look at IP behavior as well as account behavior. If you are on a VPN, corporate proxy, school network, public Wi-Fi, or mobile carrier gateway, other users may be contributing traffic from the same visible route. Turn the VPN off, switch networks, or try a trusted home connection. If ChatGPT works only when the VPN is off, use our chatgpt not working with vpn troubleshooting page.
Try another browser or the mobile app
Browser extensions can create background requests, reload pages, or interfere with ChatGPT’s session. Try a private window with extensions disabled, then a different browser. If the error appears only on iOS or Android, follow the device-specific steps in our chatgpt app not working guide.
Use a smaller prompt and avoid rapid regenerations
Long prompts, repeated file analysis, image generation, and regeneration loops can consume limits faster than short text turns. If you were using uploads, images, voice, or data analysis when the error appeared, pause that feature and test a simple text prompt. We cover nearby feature-specific failures in chatgpt file upload not working, chatgpt image generator not working, and chatgpt voice mode not working.
Why the error keeps coming back
A recurring 429 usually points to one of several patterns. Treat the error as a signal, not as a random glitch.
| Pattern | Likely cause | Best next step |
|---|---|---|
| It happens after many fast prompts | Your account or session is temporarily rate limited. | Wait, then resume at a slower pace. |
| It happens only on VPN | The VPN exit IP may have heavy shared traffic. | Turn off VPN or change to a trusted network. |
| It happens only in one browser | An extension, cached session, or tab loop may be involved. | Use a clean private window, then clear site data if needed. |
| It happens with uploads or tools | The tool may be consuming heavier resources than a normal text chat. | Try a small text-only prompt, then return to the tool later. |
| It happens across all devices | Your account may have hit a plan limit, or OpenAI may have a service issue. | Check status, wait, and review plan limits. |
| It happens in code or automation | Your app is exceeding requests, tokens, quota, or project limits. | Add backoff, queueing, token reduction, and limit monitoring. |
ChatGPT users often confuse 429 with a general outage. That can happen because both feel the same from the user side: ChatGPT stops answering. The difference is that 429 is usually about too much traffic from your side, your account, your network, or your app, while a broad outage affects many users at once. If the page loads slowly but does not show 429, our why is my chatgpt so slow guide may fit better.
Usage caps are also not always identical across accounts. OpenAI’s free-tier documentation says Free users have stricter rate limits than paid plans and may be notified when they reach a limit.[10] That does not mean every 429 is solved by upgrading. It means you should first confirm whether you are hitting a temporary limit, a feature limit, a plan limit, or a wider incident. If your main problem is message ceilings rather than an HTTP-style error, see chatgpt too many requests in 1 hour, chatgpt daily limit, and How to Bypass ChatGPT Message Limits Legitimately.
Developer fixes for OpenAI API 429 errors
If you see 429 in logs, do not solve it with blind retries. OpenAI’s error-code guide separates “rate limit reached for requests” from “exceeded your current quota,” and the fixes are not identical.[5] Rate-limit errors need pacing. Quota errors need billing, credit, or usage-limit review.
Read the error message first
OpenAI’s 429 Help Center article says the error can include the model, organization, limit, and current usage in the message.[2] Save that message in logs. It tells you whether you hit a request ceiling, a token ceiling, or another limit. Do not throw it away and replace it with a generic “try again” message.
Inspect rate-limit headers
OpenAI documents response headers for rate-limit state, including limits, remaining requests, remaining tokens, and reset timing.[3] Use those headers to control concurrency rather than guessing. If your app has multiple workers, centralize this state so every worker does not retry at once.
Respect Retry-After when it appears
The HTTP Retry-After response header tells a client how long to wait before a follow-up request, and it can be used with 429 responses.[9] If the response includes this header, use it as the minimum wait. If it does not, fall back to exponential backoff with jitter.
Add exponential backoff with jitter
OpenAI recommends exponential backoff for 429 errors: wait after the first failure, increase the wait after each repeated failure, and stop after a maximum retry count.[2] Jitter matters because it prevents all clients from retrying at the same moment. A good retry layer should distinguish 429 from authentication, validation, and server errors.
On 429:
read error type and headers
if Retry-After exists, wait at least that long
otherwise wait with exponential backoff plus jitter
retry only if the request is safe to repeat
stop after a defined retry limit
log the final failure with model, project, and request sizeReduce token demand before asking for higher limits
OpenAI’s rate-limit best-practices article recommends reducing max_completion_tokens to match expected output length and shortening prompts when possible.[4] This matters because long contexts and oversized output caps can trigger token limits even when request counts look modest. Summarize older context, remove repeated examples, and cap output length deliberately.
Queue, batch, and smooth traffic
If your app sends bursts from cron jobs, webhooks, or user imports, add a queue. Smooth the burst before it reaches OpenAI. For non-urgent work, OpenAI’s Batch API is designed for asynchronous jobs and uses a separate pool of rate limits, with a stated 24-hour turnaround target.[7] That is a better fit for large evaluations, classification jobs, and embedding backfills than synchronous retry storms.
Separate user limits from platform limits
Do not let one heavy customer consume your entire project limit. Add per-user or per-workspace quotas in your own app. OpenAI also recommends setting usage limits for individual users when you provide programmatic access or bulk processing features.[3] Your local quota should fail early with a clear message before OpenAI returns a 429.
ChatGPT 429 vs. other common errors
Several ChatGPT errors look similar because they all interrupt the conversation. The fix depends on the code and wording.
| Error | Plain-English meaning | First fix | Related guide |
|---|---|---|---|
| 429 Too Many Requests | You, your session, your network, or your app is sending more than the current limit allows. | Wait, slow down, reduce demand, or add backoff. | This guide |
| 500 Internal Server Error | The server failed while processing the request. | Retry after a short wait and check status. | chatgpt error 500 internal server |
| 502 Bad Gateway | An upstream service or gateway returned a bad response. | Refresh later and check for incidents. | ChatGPT Error 502 Bad Gateway |
| 503 Service Unavailable | The service is overloaded or temporarily unavailable. | Wait and retry after a brief pause. | chatgpt error 503 service unavailable |
| 1020 Access Denied | A security or access rule blocked the request. | Check VPN, browser, region, and network reputation. | chatgpt error 1020 access denied |
| Network error | The connection broke before ChatGPT finished responding. | Check connection stability and shorten the prompt. | chatgpt network error |
OpenAI’s own error-code guide lists 500 as a server-side processing error and 503 as overload or high traffic, while 429 is tied to request rate or quota.[5] That distinction matters. A 500 or 503 often calls for patience. A 429 calls for slower behavior from the client.
When to wait and when to escalate
Wait if the error appeared after a burst of normal use, if OpenAI’s status page shows a related incident, or if your app logs show a temporary request or token spike. In those cases, the best fix is pacing. Do not make the system prove the same point by retrying continuously.
Escalate if the error persists across devices, networks, and browsers after a reasonable break; if you are a paying workspace admin and all members are affected; or if API logs show you are under documented limits but still receiving repeated 429 responses. OpenAI’s rate-limit best-practices article notes that quantized enforcement can cause bursts to fail even when a per-minute total appears below the headline limit.[4] That means your logs should include timestamps fine-grained enough to reveal short spikes.
For API teams, escalation should include the organization, project, model, timestamps, request IDs if available, error body, headers, and observed traffic rate. Also include what you already changed: backoff, reduced token caps, queueing, and user-level throttles. This prevents a support loop where the first answer is simply “slow down requests.”
For ChatGPT users, escalation should be simpler. Note the exact message, browser, device, approximate time, whether VPN was enabled, and whether the same account works elsewhere. If the issue is actually a missing conversation, a stuck response, or a generic banner, use chatgpt conversation not found error, chatgpt error in message stream, or chatgpt something went wrong error instead.
Frequently asked questions
Is ChatGPT error 429 my fault?
Not always. It can be caused by your own rapid prompts, a shared network, browser behavior, a plan limit, or an OpenAI-side load condition. The right response is to pause first, then narrow the cause by changing one variable at a time.
How long should I wait after a 429 error?
If ChatGPT gives a retry time, follow it. If it does not, wait a few minutes before trying again, and do not spam the send button. API clients should use Retry-After when present and exponential backoff when it is not.
Will upgrading ChatGPT fix error 429?
Sometimes, but not always. Paid plans generally have higher usage limits than Free, but a VPN issue, browser loop, service incident, or API bug can still produce errors. Confirm the cause before paying only to solve a temporary rate limit.
Why do I get 429 when I am below my OpenAI API limit?
Your traffic may be bursty inside a smaller time slice, or you may be hitting a token limit rather than a request limit. OpenAI’s guidance says limits can be quantized over shorter periods, so smooth traffic instead of only watching minute totals.[4]
Should I keep retrying until ChatGPT works?
No. Repeated retries are the common mistake. OpenAI says unsuccessful requests can still count against the per-minute limit, so continuous resending can keep you limited.[2] Wait, then retry once at a slower pace.
Can a VPN cause ChatGPT 429?
Yes. A VPN or proxy can put many users behind the same visible IP address, which may look like high-volume traffic. Turn it off temporarily, switch to a trusted network, and test again before assuming your account is broken.