Limits
Fair-use free tier. No SLA. Limits may change. Best effort.
| Dimension | Value | On exceed |
|---|---|---|
| Requests / minute | Elastic — no fixed per-account RPM cap | 429 + Retry-After |
| Requests / day | 10,000 successful / account (UTC) | 429 |
| Concurrent in-flight | Elastic — up to 3 while idle; reduced as demand increases | 429 |
| Input cap / request | 8,000 model tokens (state + all questions), max 32,000 UTF-8 bytes | 413 |
| API keys / account | 3 | Create disabled in Console |
Requests are admitted while shared service capacity is available. Busy service capacity returns 429 with Retry-After. Daily and concurrent limits are per account and shared by all of its keys (up to 3, for rotation). Daily quota counts successful inference only; rejected, failed or timed-out requests are not charged.
Best practices
- Backoff on
429/504/5xx. - Keep
state+ questions under the soft cap. - Concurrency is shared by all keys of an account. Idle capacity allows up to 3 concurrent requests; competing accounts and busier capacity reduce that share to 2 or 1. Existing requests finish normally. Retry on 429 with backoff.
- Do not resell quota or scrape the free tier.