# Limits

Limits depend on the key, pool, model and request format. Do not assume one global context window or concurrency allowance.

Source: https://zrelay.net/docs/api/limits/

## Model limits

Check the selected route’s context and output limits before sending a large request. Bound your output budget explicitly. Prompt size includes system messages, tools, conversation history and relevant cached context.

## Account limits

Per-key spending caps, account balance and route eligibility are independent checks. A valid key with available account balance can still be blocked by its spending cap or model scope.

## Traffic and concurrency

Handle 429 responses and respect Retry-After when provided. Use bounded concurrency, request timeouts and a retry budget. No fixed requests-per-minute number is promised across all pools by this documentation.

## Set useful controls

See [Spending controls](https://zrelay.net/docs/md/capabilities/spending-controls.md) and [Errors](https://zrelay.net/docs/md/api/errors.md). For large workloads, contact support with your model, pool and expected concurrency before depending on a particular capacity.