openaienmodel: gpt-5-mini-2025-08-07
AI Gateway - Automatically retry on upstream provider failures
Key Points
- Gateway-level automatic retries
- Configurable attempts, delay, and backoff
- Per-request overrides; use Dynamic Routing for multi-provider failover
Summary
AI Gateway now performs automatic retries for failed upstream provider requests at the gateway level. This lets you apply retry logic centrally (without changing clients) and handle transient upstream errors before they reach callers.
Key Points
- Gateway-level retries configurable globally and overridden per request via headers.
- Retry options:
- count: up to 5 attempts
- delay: 100ms to 5s
- backoff: Constant, Linear, or Exponential
- Defaults apply to all gateway requests unless a per-request header override is provided.
- Use this when clients cannot implement retry logic; for cross-provider failover or more advanced routing, use Dynamic Routing.
- See Manage gateways for full configuration and API details.