Tiers
Every account starts on the Core tier.
The figures apply to each model separately. The Tier and limits page on the platform shows the figures for the account. An account with individual limits shows the Custom tier.
Requests through OpenRouter or another router follow that router’s limits.
Moving up
- What counts: usage charged to purchased credits in one calendar month (UTC). The amount of credits bought does not count, and neither do promotional credits.
- When: any month that reaches the figure counts, including the current one.
- How: press Request upgrade on Tier and limits or on the Account page — see Get help. Each request is reviewed. A new tier starts at the beginning of the next month at the earliest.
- Keeping it: a tier stays with the account. There is no downgrade.
- More than Scale: in the same form, choose Enterprise API terms. For on-premises or private cloud, see Harbor.
How requests are counted
- The windows are rolling: the last 60 seconds and the last 60 minutes, not clock minutes.
- All keys of the account share the limits. Each model has its own count.
- Every admitted request counts, also one that fails later. A request refused at admission does not count.
- Over a limit, a request is refused with
429 rate_limit_exceededand aRetry-Afterheader — see Errors.
Requests running at the same time
A separate limit caps how many requests from one account can run at once for each model, across all keys. It is not published and can change with the load. A request over it fails at once with429 concurrency_limit_exceeded.
The Terms allow Fortytwo to throttle or queue requests for a short time to protect the service; a request can then wait before it starts — see Terms of Service, section 2.
Request size
Image limits are on Image input. An image sent as a data URL counts toward the request body.
How long a request may run
A request can run for up to 60 minutes. After that a non-streaming request returns504 upstream_timeout; a stream ends with an upstream_error event — see Streaming.