Every plan carries throughput ceilings that the Gateway enforces per API key: a request rate, a concurrency cap, and a token budget per minute counted on input and output separately. They bound instantaneous load; your plan’s monthly allowances still govern total usage.
Limits by plan
| Plan | Requests per minute | Concurrent requests | Input tokens per minute | Output tokens per minute |
|---|---|---|---|---|
| Starter | 100 | 5 | 2,000,000 | 2,000,000 |
| Business | 100 | 15 | 2,000,000 | 2,000,000 |
| Advanced | 300 | 45 | 6,000,000 | 6,000,000 |
Enterprise agreements set their own ceilings as part of the contract.
When a limit is hit
A throttled request is rejected immediately with 429 — rate-limit-exceeded or, when the concurrent-request cap is reached, *-concurrency-exceeded. The Gateway does not queue over-limit traffic; back off briefly with jitter and retry, as covered in Errors and retries.