Source: OpenAI — Flex processing guide

OpenAI's `service_tier: "flex"`: Batch-rate pricing on synchronous calls in exchange for slower responses and occasional 429 Resource Unavailable, with longer timeouts and fallback guidance.

Contents

Summary

Flex is a cheaper synchronous tier for evals, enrichment and async jobs. It is billed at Batch API rates but may be slow or temporarily unavailable.

Key claims

  • Enabled per request with service_tier="flex" on Responses or Chat Completions; beta, limited model support.
  • Pricing equals Batch API rates; prompt caching discounts stack.
  • Official SDKs default to a 10-minute timeout; the guide suggests raising it (examples use 15 minutes) for flex. SDKs retry 408 Request Timeout twice.
  • When capacity is short, flex returns 429 Resource Unavailable, which is not billed.
  • Two retry strategies: exponential backoff when cost matters more than latency, or fall back to the default tier (service_tier="auto" or omit) when completion matters more.

Feeds

Rate limits and retries

Suggest changes →