Pricing & usage
Paid plans have a daily request-unit allowance, a daily request-count guard, and a requests-per-minute (RPM) limit. These controls measure different things. Exact plan amounts, prices, and model tiers are maintained on Plans; this page explains how the usage controls work.
The three usage controls
| Control | What it measures | When it is exhausted |
|---|---|---|
| Request units | Weighted model usage based on model cost and request shape | New billable usage is blocked until the daily reset or eligible prepaid fallback |
| Request-count guard | Number of Gateway API requests during the daily window | Further API requests are rejected until reset |
| RPM | Short-term request rate | Requests are temporarily rate limited; retry after the reset window |
Request units reset at midnight UTC and do not roll over. They are plan usage capacity, not money, and cannot be transferred or redeemed for cash.
How request units are calculated
One API call does not always equal one request unit. The amount depends on the configured model price and usage reported for the request. Input and output tokens, cached tokens, audio, generated characters, and other supported usage components may be priced differently. Large contexts, long outputs, tools, and multimodal input can therefore use more units than a short text request.
The Model Hub shows an estimate where available. Treat it as a planning aid: the final amount follows the usage reported by the selected provider model and Accly’s current pricing configuration.
Model and account access
Usage capacity does not grant access to every model. API-key access, plan tier, the key’s model allowlist, model status, and endpoint protocol are checked separately. A model or route can also be unavailable while its provider is degraded; Accly does not silently substitute an unrelated cheaper model.
See Plans for tier access, Authentication for key controls, and Models for protocol and model selection.
Prepaid fallback
An eligible paid account can opt to use prepaid credits after its daily allowance is exhausted. The setting does not bypass the plan’s model access, RPM, or daily request-count guard. Read Prepaid Credits for pack values, activation, settlement, and purchase records.
When a request is blocked
The Gateway returns a structured error with an error.code. Use that code to distinguish rate limiting, exhausted daily capacity, model access, and account-state failures. Follow the retry guidance and exact error behavior in Errors & Limits; do not retry a rejected request in a tight loop.
The hosted Free trial is a separate one-time app allowance. It does not reset as a daily Gateway API quota, and Free accounts cannot create API keys. See Plans for the Free account behavior.