Limits
These ceilings are shared, not allocated per key.
| Limit | Value |
|---|---|
| Requests per minute, per key | 120 |
| Requests per minute, per source IP | 300 |
| Identical request repeated | more than 3 times in 10s → rejected |
These ceilings are shared, so sustained traffic from anyone on the platform counts against the same limit. Treat these as an upper bound you should stay well under, not as a reservation.
The per-IP ceiling usually binds first: everyone behind one address shares 300 requests per minute, which is 5 per second. The duplicate rule counts byte-identical requests, so a retry that changes nothing is rejected rather than forwarded — vary the request or wait. A rejected duplicate is not billed.
Balance
Each key carries a prepaid balance. When it is exhausted, requests are rejected until it is topped up. A streaming request already in flight when the balance runs out is allowed to finish, so the final charge can land slightly past zero.