API REFERENCE
Rate limits
What limits apply to each kind of request, and how a limit is reported.
Z1, Z1 Code and H1
The commercial families have no published requests-per-minute limit. A request is bounded by your wallet balance, and each request by a per-request ceiling: max_tokens above 16384 is refused with 413 request_too_large before anything is spent. When the wallet cannot cover a request, the answer is 402 insufficient_credits.
Project ZERO Alpha
| Limit | Value |
|---|---|
| Requests per developer per UTC day | 200 |
| Requests per developer per minute | 30 |
| Concurrent requests | Limited; not published as a number |
| Spend allowance | A per-developer daily ceiling and a shared program ceiling |
A ZERO limit is reported as HTTP 503 zero_capacity_unavailable (not 429), with error.limit (rate, concurrency, requests, daily, global or provider_capacity), error.reset_at and a Retry-After header in seconds. Per-minute limits clear at the next minute; daily limits at 00:00 UTC.
Embeddings
64 inputs per request. No other limit is published.
Good client behaviour
- Honour
Retry-Afterandreset_at; do not retry a refused request in a tight loop. - Back off exponentially on 502.
- Read
GET /v1/zero/statusbefore a ZERO batch. - Treat the numbers on this page as Alpha values that can change.