OpenAI has added two discounted completion windows to background mode in the Responses API, now in public beta. Setting background_options.completion_window to 2h cuts model token rates 30% below Standard, and 8h cuts them 40%; the API targets returning the first response within the chosen window after it accepts the request. The windows work on GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna. The OpenAI Developers X account promoted it on November 18; the API changelog dates the launch to November 17, the same day as GPT-6.2.

Key Takeaways

  • ✓Discounts: model token rates are 30% below Standard with the 2h window and 40% below with the 8h window; the default soonest window runs on demand with no discount
  • ✓Models: GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna, on v1/responses only
  • ✓Usage: set background to true and background_options.completion_window to 2h or 8h, leave stream unset or false, and use Standard processing (service_tier default)
  • ✓Limits: 2h and 8h requests share the Batch/Background rate limit with Batch API jobs, and regular Standard requests cannot reuse prompt cache entries created by these background requests
  • ✓Multi-turn: a continuation_id links follow-up requests, which also get the discount and higher queue priority
⚡

Key Decision Metrics at a Glance

TL;DR Verdict
🧭

Turn your technical choice into a development budget

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

🔬

In-Depth Technical Analysis

Background

Many agent and data jobs do not need an instant answer, such as reviewing documents overnight or analyzing the morning's support tickets before an afternoon meeting. Until now the main way to save money was the Batch API, which fits poorly with tool-calling, multi-turn workflows. OpenAI has now built the discount directly into background mode in the Responses API.

How it works

Background mode gains a background_options.completion_window field with three values: soonest (the default, no discount), 2h and 8h. The 2h and 8h windows are targets rather than guarantees: after accepting a request, the API aims to return the first response within the window and automatically retries eligible capacity errors. Multi-turn jobs can pass a continuation_id so follow-up requests keep the discount and get higher queue priority.

Numbers that matter

There are no benchmarks here; the story is price. Per OpenAI's docs and changelog, the 2h window cuts model token rates 30% below Standard and the 8h window 40%. At GPT-6.2 Astra's Standard price of $8 per million input tokens and $40 per million output tokens, the 8h window works out to $4.80 and $24.

What it means for developers

Set background: true and completion_window: "8h", leave streaming off and set service_tier to default, then fetch the result by webhook or by polling the response ID. Note that 2h and 8h requests count against the Batch/Background rate limit, and the prompt cache they create cannot be reused by regular real-time requests.

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

Up NextChatGPT Dots Now Follow Custom Instructions: Set Tone and Which Actions Need Your Go-Ahead in the Thread, Plus Faster Replies, Less Jargon and Better Follow-Through
Read Now→