> ## Documentation Index
> Fetch the complete documentation index at: https://docs.worldlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Rate limits

> Rate limits and throughput options for World API world generation

## Rate limits

World API rate limits protect capacity and keep world generation available for all users. The limits on this page apply to requests that start a world generation with [`POST /marble/v1/worlds:generate`](/api/reference/worlds/generate).

There are two relevant types of limits:

1. **Start-rate limits** control how many world generation requests your API user can start over a time window.
2. **Throughput tiers** determine whether your account uses the default limits or an approved higher-throughput configuration.

## Current limits

| Tier                                 | Requests covered                 | Limit                                                |
| ------------------------------------ | -------------------------------- | ---------------------------------------------------- |
| Default                              | World generation starts          | About 3 requests per minute and 60 requests per hour |
| Higher-throughput, approved accounts | Standard world generation starts | About 30 requests per minute                         |
| Higher-throughput, approved accounts | Draft world generation starts    | About 90 requests per minute                         |

Standard world generation means non-draft models such as `marble-1.0`, `marble-1.1`, and `marble-1.1-plus`. Draft world generation means `marble-1.0-draft`.

Limits are enforced for the API user or account associated with the request, not for each API key.

## How limits are applied

* The limits on this page apply to generation starts, not to the number of generations already running.
* Usage is tracked in rolling windows, and enforcement is approximate. Short bursts can hit a limit even when average usage over the full window appears lower.
* These limits are maximum allowed usage, not guaranteed minimum throughput. Available capacity can vary with service load.
* Each accepted world generation usually takes about 5 minutes to complete. Use [`GET /marble/v1/operations/{operation_id}`](/api/reference/operations/get) to poll for the result.

## 429 errors

If you exceed a rate limit, the API returns `429 Too Many Requests`. Treat a `429` as a signal to slow generation starts for that API user or account.

When handling `429` responses:

1. Pause before retrying. If the response includes `Retry-After`, wait at least that long.
2. Retry with exponential backoff and jitter.
3. Avoid retrying many queued starts at once after a pause.
4. Reduce concurrent start attempts if you continue seeing `429` responses.

A generation request that returns `429` has not been accepted. After retrying successfully, the response will contain an Operation object; poll that operation instead of resubmitting the same start request.

## Requesting higher limits

If your workload needs sustained concurrent generation starts, contact [support@worldlabs.ai](mailto:support@worldlabs.ai) or your World Labs account representative with:

* The API user or account email.
* The models you plan to use.
* Expected starts per minute and starts per hour.
* Whether traffic is bursty or sustained.
* A short description of the customer workflow or launch timeline.

Higher limits are reviewed per account and may use separate standard and draft world generation limits.