The fastest queue is unnecessary for a report that no one will read until tomorrow. But a lower token price does not automatically make the entire automation cheaper.

What actually happened

OpenAI introduced completion windows for Responses API background mode in public beta. Eligible model tokens are priced 30% below Standard when a request uses a two-hour completion window and 40% below Standard with an eight-hour window. Supported models include GPT-6.2 Astra and Sol, GPT-6.1 Terra and GPT-6 Luna.

The request needs background processing enabled and background_options.completion_window set to '2h' or '8h'. Set service_tier to 'default' for predictable Standard-tier eligibility. The selected window targets the initial response after acceptance; multi-step jobs still have to handle any required follow-up calls.

This discount is on eligible model token rates, not every cost in a pipeline. It does not remove the need to handle failed responses or duplicate callbacks.

Why a small team cares

Imagine a six-person agency using n8n to categorize support requests and prepare a morning issue summary. The job runs after business hours, and a two-minute response has no more value than a response delivered overnight, provided it arrives before the team starts work.

Document processing, enrichment and scheduled summaries are similar candidates. Customer-facing chat, checkout steps and live operator tasks are not.

Using the lower-priced mode may be a small code change, but dependable scheduling still requires stored response IDs, completion notifications and proper retry handling. If the cheaper request misses a deadline or triggers duplicate runs, the apparent savings disappear.

Hype vs useful

A 40% reduction in eligible token rates does not mean a 40% reduction in the bill for the workflow. External calls, infrastructure, retry frequency and model choice also count.

The Batch API may be preferable for independent inputs. Older models with Flex support can sometimes be cheaper; GPT-6.2 does not support Flex according to the current developer discussion. Benchmark the full cost per completed job at your actual volume.

Move one overnight workflow first, compare its reliability and spend with the current setting, and expand only if the result survives that comparison.