In plain words
Every AI service runs on a limited number of expensive computers shared by millions of people. To keep things fair, it sets a cap on how much each person or app can ask for in a given time: so many messages, or so many tokens, per minute, hour or day. That cap is a rate limit.
Go over it and nothing breaks. The service simply says no for now: "too many requests", "you've reached your limit" or "try again later". Wait a little and you can carry on.
In chat apps, the limit usually shows up as a message cap that resets after some hours, with higher caps on paid plans. For developers using an API, limits are written down per minute and per day, and they grow as an account builds up a history of use and payment.
Everyday example
The ATM's daily withdrawal limit
Your bank card lets you take out only so much cash a day, and the ATM refuses more until tomorrow. The money is still in your account. The limit protects the system, not your balance. An AI service does the same with requests: it serves you up to the cap, then politely refuses until the window resets.
Where the example breaks: An ATM limit is one fixed number per day. AI rate limits often refill gradually and count several things at once, such as requests and tokens, so you can hit one even while the others are fine.
How it actually works
OpenAI's documentation sums up the purpose: limits "help protect against abuse or misuse of the API" and help "ensure that everyone has fair access".
Several things are counted
AI APIs typically measure requests per minute, tokens per minute and sometimes requests per day or images per minute. Whichever cap you reach first is the one that stops you. Limits are usually set per account or project and differ by model.
The bucket refills
Some providers use a "token bucket": your allowance refills continuously rather than resetting on the hour. Anthropic's docs note that 60 requests per minute may be enforced as about one per second, so a sudden burst can be refused even if the minute's total looks fine.
The refusal has a number: 429
Over the limit, a web service replies with HTTP status 429 Too Many Requests, defined in 2012. Often it adds a
retry-aftervalue saying how many seconds to wait.Apps retry politely
Well-built apps wait and try again, waiting longer after each failure (called exponential backoff) and adding a little randomness so thousands of clients don't all retry at the same moment. Failed attempts can still count against the limit, so hammering the service only makes it worse.
Tiers raise the ceiling
Providers place accounts in usage tiers. As an account uses and pays for more, it moves up automatically and gets higher limits. A separate spend limit can cap the monthly bill.
- Error code
- HTTP 429 Too Many Requests, defined in RFC 6585 (April 2012)
- Common units
- Requests per minute (RPM), tokens per minute (TPM), requests per day (RPD)
- Why they exist
- Stop abuse, share limited computing fairly, and protect the service from overload
- Example reset
- Google's Gemini API resets daily request quotas at midnight Pacific time (as of November 2026)
Exact caps for each plan and model change often, so we don't list them here. Check the provider's own limits page.
Questions people ask
Why does ChatGPT say I've reached my limit?
Chat apps cap how many messages, or how much heavy work like long files and advanced models, each plan can use in a time window. When you hit it, you wait for the reset or switch to a lighter model. Paid plans raise the cap but still have one.
What does error 429 mean?
It is the web's standard reply for "too many requests". The service is working but you, or the app you are using, asked too often in a short time. Waiting, then retrying more slowly, fixes it.
Is a rate limit the same as a usage limit or a token limit?
Close cousins. A rate limit caps how fast you ask. A usage or spend limit caps how much in total, over a day or a month. A token limit usually means the context window: how much text fits in one conversation.
Do rate limits reset?
Yes. Per-minute limits refill within seconds to a minute. Daily quotas reset once a day, at a time the provider sets. Chat app caps reset after a stated period, which varies by plan and service.
Can I get a higher rate limit?
On chat apps, a paid plan usually gives a higher cap. On APIs, limits rise automatically through usage tiers as an account builds history, and providers offer forms or sales contacts for more.
Was this clear?
1 is confusing, 5 is crystal clear. One tap, no account needed.
Spotted something wrong, or know a better example? Suggest a change