Definition
Rate limiting is Rate limiting is a cap on how many requests a user, app or agent may make to a service in a set time, used to prevent overload and abuse and enforced by AI APIs and websites alike.
Rate limiting is a rule that caps how often you can ask a service for something. It might say "60 requests per minute" or "100,000 tokens per minute". Go over the line and the service slows you down, returns an error, or blocks you for a while.
There are three main reasons.
AI providers usually set limits in more than one unit: requests per minute, tokens per minute, and sometimes tokens per day. Your limit often rises as you spend more or use the service longer. If you hit it, the API typically returns an error code, and your code should wait and retry with growing delays instead of hammering the service. An AI gateway can queue requests or fall back to another provider when one is limited.
Limits also tie to your API key. A leaked key is dangerous partly because it spends your quota and your money.
Websites apply the same idea to visitors. Social platforms and online stores watch the pace of actions on an account. A browser agent that reads pages or clicks far faster than a person can look like a bot, and the platform may slow it, ask for verification, or suspend the account. Many sites also set rules for web crawlers, and a file called robots.txt states them, though it only works if the crawler follows it.
If you build with AI, plan for limits from the start. Add retries with backoff, spread work over time, and watch your usage. If you run an agent inside accounts you care about, slow it to a human-like pace, use an official API where one exists, and keep a separate account for heavy reading. Treat a rate limit as a boundary the service has set, not a bug to route around.
Rate limiting caps how many requests a user or app can send to a service in a set period, to prevent overload, abuse and runaway cost.
Wait and retry with increasing delays, spread requests over time, and ask the provider about a higher limit if you need steady volume.
OpenAI said Monday it will add an invisible watermark to eligible ChatGPT and Codex text generated in the EU over the coming weeks, and API customers anywhere can now opt in. It is a compliance signal, not a lie detector, and OpenAI says so itself.
Anthropic expanded Claude for Startups on Tuesday: eligible startups get a free year of Claude Team for up to five Premium seats, plus $1,000 in API credits. It is a bid to be your default by year three, and still a fair deal if you read the fine print.
Google released Nano Banana 2.1 on October 6, an update to its Flash-tier image model that, by Google's own evaluations, holds multi-character scenes together far better. If you make images with people in them or infographics with real text, test it this week.
OpenAI put its Decisions API in public beta on October 6, with gpt-6-luna at $0.10 per million input tokens and no charge for output. If you classify, route or grade at volume through a chat model, this is the endpoint to price against.
Anthropic told prospective investors its third-quarter revenue reached $16.8 billion, up from about $11.5 billion, and that it was profitable on an adjusted basis, the FT reports. Growth is slowing. The profit is the bigger news for anyone who pays Anthropic.