Limits

How requests are limited, and how you find out.

Rate limits

Each key has a requests-per-minute limit: the platform default, or the number you set on the key. Every response tells you where you stand:

Header

Meaning

X-RateLimit-Limit

Requests allowed per minute.

X-RateLimit-Remaining

Requests left in the current minute.

Retry-After

Seconds to wait, sent with a 429 .

The default limits are published with the API launch and shown on each key in the console.

Spend caps

•

A key cap stops that key for the rest of the calendar month. Other keys carry on.

•

The organization limit under Billing stops all its keys together.

•

A key's model list rejects any other model with a 403 .

A request refused by a cap returns 402 and is not charged.

Handling 429 and 5xx

Retry only when the failure may be temporary ( 429 and 5xx ), wait longer each time, and give up after a few tries:

JavaScript

Python

async function call(url, init, tries = 5) {
  for (let attempt = 0; attempt < tries; attempt++) {
    const response = await fetch(url, init);
    if (response.status !== 429 && response.status < 500) return response;
    const wait = Number(response.headers.get("retry-after")) || 2 ** attempt;
    await new Promise((r) => setTimeout(r, wait * 1000));
  }
  throw new Error("Still failing after retries");
}

For model requests you often don't need to retry yourself: set fallback models and the request moves on by itself.

Usage and billing

Errors