Rate limits & restrictions
There are no artificial token quotas, daily caps or model gating — you pay for the tokens you use. The limits that do exist protect the gateway, and normal coding-agent traffic never hits them.
| Limit | Value | When exceeded |
|---|---|---|
| Requests per minute | 600 / min per client | 429 — back off and retry |
| Request body size | 25 MB | 413 — trim attachments / context |
| Gateway timeout | 180 s per request | 504 upstream_timeout — use streaming |
| Token limits | none of our own | only the model's native context window applies |
Streaming and long generations
The 180-second timeout applies to how long the connection can sit without completing. With "stream": true tokens flow as they're generated, so even very long outputs finish fine. Every tool documented in the setup guides streams by default — the timeout mostly bites hand-written non-streaming scripts with large max_tokens.
Concurrency
There is no artificial concurrency cap on the OpusGate side; parallel requests are bounded only by upstream model throughput. If you hit sustained upstream 429s at high parallelism, add exponential backoff — or contact support to discuss your workload.
Error bodies for 429 / 504 and friends are on the error codes page.