What changed

The OpenAI Responses API now offers two-hour and eight-hour background completion windows in public beta for supported models. Token rates are 30% below Standard for the two-hour window and 40% below Standard for the eight-hour window.

Why it matters

Long-running agents can trade completion latency for lower token rates without moving work to a separate batch interface, but the longer execution and retention window changes timeout, cancellation, idempotency and cost-control assumptions.

What to do

Replay a non-production long task through Standard, two-hour and eight-hour modes; measure total cost and wall time, interrupt each run, retry with the same idempotency key and verify final external state before adoption.

Caveat

This is a public beta and OpenAI’s changelog does not establish completion-time guarantees, workload eligibility, queue behavior or recovery semantics for every failure mode.