Free guide
OpenAI Background Windows: GPT-6.2 at 30% to 40% Less If It Can Wait
Same model, lower token rate, if your job can wait. How the 2-hour and 8-hour windows work, how to set one up, and the fine print.

The short version
- Set background to true and background_options.completion_window to 2h (30% below Standard) or 8h (40% below Standard). The default, soonest, has no discount.
- It's the same model at a lower token rate. The trade is time: the API targets the first response within the window you picked.
- Follow-up turns on the same window get the discount too. No streaming on 2h or 8h, it needs Standard processing, and it's a public beta.
How it actually works
Like a photo drop-off
One-hour prints cost the most. Tick a later box on the envelope and the price drops. Background windows work the same way: soonest is the default with no discount, 2h is 30% below Standard and 8h is 40% below Standard. OpenAI doesn't say why it can charge less, only that you're giving it time.
You get a claim stub right away
A background request returns a response ID as soon as the API accepts it. The work happens on OpenAI's side. While the response is queued or in_progress, you either wait for a webhook (subscribe to response.completed) or poll GET /v1/responses/{id}.
The window is a target for the first response
The completion target starts when the API accepts the request and applies to that first response. OpenAI's docs say start times depend on scheduling, and that after accepting a 2h or 8h request the API automatically retries eligible capacity errors. The docs don't say what happens if a window is missed.
Follow-ups keep the discount
The discount applies to model token rates on each eligible request, including follow-ups. Create the follow-up with the same model, service tier, background setting, completion window and continuation ID, and set previous_response_id to the earlier response's ID. That's how a multi-turn agent run stays on the cheaper rate.
How to set it up
1.Pick a job that can wait
Nightly evals, report drafts, morning briefs, summaries you read tomorrow. Anything a user is watching live (a chatbot, a coding agent you're steering) should stay on the default window, because 2h and 8h requests can't stream.
2.Send it with an 8h window
Background mode on, Standard processing (service_tier default), and the window inside background_options. Save the ID that comes back.
curl https://api.openai.com/v1/responses \ -H "Authorization: Bearer $OPENAI_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-6.2-sol", "input": "Summarize yesterday'\''s support tickets by theme.", "background": true, "service_tier": "default", "background_options": { "completion_window": "8h" } }'3.Get the result: webhook or poll
Subscribe a webhook to response.completed (and the failed, incomplete and canceled events you care about), then fetch the response with event.data.id. Or poll the ID while the status is queued or in_progress.
curl https://api.openai.com/v1/responses/resp_123 \ -H "Authorization: Bearer $OPENAI_API_KEY"4.Send follow-ups on the same window
Keep the same model, service tier, background setting, completion window and continuation ID, and set previous_response_id to the last response's ID. If the earlier response has expired, leave previous_response_id out and pass the history in input.
5.Cancel if you change your mind
Canceling twice is safe (it's idempotent).
curl -X POST https://api.openai.com/v1/responses/resp_123/cancel \ -H "Authorization: Bearer $OPENAI_API_KEY"
The fine print
- Public beta, on GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna only.
- The discount is on model token rates: 30% below Standard for 2h, 40% below for 8h. Standard GPT-6.2 Sol is $2 in and $10 out per million tokens, so 8h works out to $1.20 and $6 (our math, not an OpenAI table). Astra is $8 and $40 at Standard, so $4.80 and $24 at 8h (our math).
- Needs Standard processing: set service_tier to default. With auto, it only qualifies if the API picks Standard. GPT-6.2 Astra and Sol don't support Flex processing at all.
- No streaming on 2h or 8h. To stream a background response, use the default soonest window.
- 2h and 8h requests share the Batch/Background rate limit with Batch API jobs, and input tokens count when you submit.
- Standard requests can't reuse prompt cache entries created by 2h or 8h requests, even for an identical prompt.
- Background mode stores response data temporarily, even with store set to false or Zero Data Retention. OpenAI says it's encrypted at rest.
FAQ
How much cheaper are OpenAI's background completion windows?
Model token rates are 30% below Standard with a 2h window and 40% below Standard with an 8h window. The default soonest window has no discount.
Is it a cheaper or smaller model?
No. You pick the same model (for example gpt-6.2-sol). Only the token rate changes, in exchange for letting the first response take up to the window you chose.
Do follow-up turns get the discount?
Yes. OpenAI's docs say the discount applies to each eligible request, including follow-ups, when they use the same model, service tier, background setting, completion window and continuation ID.
Can I stream a 2h or 8h request?
No. For 2h and 8h, leave stream unset or false. Streaming background responses only works on the default soonest window.
Is this the same as Flex processing or the Batch API?
No. GPT-6.2 Astra and Sol don't support Flex. Background windows are part of the Responses API and need Standard processing. They only share a rate limit with the Batch API.
Sources
- OpenAI Devs on X: discounted windows in public beta
- OpenAI docs: Background mode and completion windows
- OpenAI API pricing
- OpenAI API changelog (Nov 17, 2026)
- OpenAI docs: Using GPT-6.2
2026-11-19 23:30 PT against OpenAI's background mode guide, API pricing page, API changelog (Nov 17), the Using GPT-6.2 guide and @OpenAIDevs' November 18 post.
Want help building with tools like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolMore free guides
Chrome DevTools MCP: Let Your Coding Agent See the Page It Builds
Google's free, open-source way to give your coding agent a real Chrome window and the DevTools you already use. What it does, how it works, how to set it up, and the fine print.
GPT-6.2 Drops Its Reasoning When You Edit Old Messages: How to Keep It
Every reasoning item is now tied to the conversation before it. Edit that history and the API drops the reasoning that follows. Here's how it works and how to keep it.
Inkling 2 Mini: Run Thinking Machines' Looped AI Model on Your Mac
Thinking Machines' new open-weights model runs on a 32GB Mac with the wifi off. What it is, how the loop trick works, how to set it up, and the fine print.