Limits and billing
Rate limits, how a request is priced and charged, and the credit limits that return 402.
Default limits, how a request is priced and charged, and the credit limits that return 402.
Rate limits
| Item | Default |
|---|---|
| Model requests per account | 600 per minute |
| Successful model requests per account | 600 per minute |
| Per API key | 300 per minute |
| Per source IP | 1200 per minute |
| Requests in progress per account | 20 |
| Async task submissions per account (coming soon) | 60 per minute, 20 in progress |
| Request body | 20 MB |
| Longest request | 600 seconds |
| Non-streaming wait for a response | 95 seconds, then 500 |
These are defaults; Relayvia may adjust them for individual accounts. Going over returns 429 rate_limited with Retry-After, and requests refused by a rate limit are not charged.
Credits
Relayvia is prepaid. Your account tops up credits in US dollars, and every team member and API key of the account spends from them. The console header and /v1/balance show what is available.
How requests are priced
Token-priced models charge per million input, output, cache-read and cache-write tokens; reasoning tokens count as output. Some models charge per call instead. Each model's price is on the pricing page and its model page, and your key's group can apply a discount. A failed request that produced no output is free.
Reserve, then settle
Before a request is sent, Relayvia reserves an estimate: the input size plus your output limit (max_tokens, max_completion_tokens, max_output_tokens or maxOutputTokens, capped at 32,768 tokens by default). If the credits or a limit cannot cover it, the request returns 402 without being sent. When the response ends, the reservation is released and the actual usage is charged, which can be more or less than the estimate. A very large output limit can cause a 402 even with credits left; lower it if that happens.
If you disconnect
Closing the connection does not cancel work already sent to the model. Relayvia keeps reading for up to 30 seconds to get the final usage and charges it; if none arrives, it charges the input plus an estimate of the output. Requests that fail before producing any output are free.
Credit limits
Besides the account's credits there are three limits: the API key's credit limit (total, daily, weekly or monthly), the team member's credit limit, and account spend limits. Running out of any of them returns 402 insufficient_quota. When the account's credits go below zero, the account is frozen until it is topped up.
Data policy
Relayvia does not store prompts or responses; usage logs only keep the model, tokens, credits, latency and status.