Rate Limit
A cap on requests/tokens per minute or concurrent requests enforced per key and tier.
- Dimensions: RPM, input TPM, output TPM, concurrency.
- Hitting cap returns HTTP 429 with retry-after.
- Higher tiers unlock after sustained spend.
When to use: In production systems: implement exponential backoff with jitter on 429s.
Example: Anthropic Tier 1: 50 RPM and 50K input TPM for Sonnet; SDK retries automatically with backoff.