Last updated: November 18, 2026
OpenAI background mode now lets you pick a completion window. Since November 17, 2026, Responses API requests sent with background set to true can choose a two-hour window at 30% below Standard token rates or an eight-hour window at 40% below. The public beta covers GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna.
Contents
- What changed on November 17
- Completion window pricing
- How to request a discounted window
- Multi-step agent runs and continuation IDs
- Background windows vs Batch and Flex
- Limits and caveats
- Frequently asked questions
What changed on November 17
Background mode already let you run long Responses API requests asynchronously. What OpenAI added, according to its API changelog entry dated November 17, is a choice of “a two-hour or eight-hour background completion window in the Responses API public beta.”
The setting lives in a new field, background_options.completion_window, described in the background mode guide. It accepts three values:
- soonest (default): no discount. OpenAI’s example is an on-demand report while someone waits for the result.
- 2h: 30% below Standard. The example is analyzing the morning’s support tickets before an afternoon triage meeting.
- 8h: 40% below Standard. The example is reviewing a batch of documents overnight for the next morning.
For the 2h and 8h options, the guide says the API “targets completion within the selected window after it accepts the request.” The target starts when the API accepts the initial request. The guide also documents an optional continuation_id field, covered below, for linking follow-up requests in one workflow.
The discounted windows arrived the same day as GPT-6.2 Astra and Sol. For background on those models, see our GPT-6.2 Astra and Sol guide.
Completion window pricing
OpenAI publishes the discounts as percentages, not as separate price lists. The table applies those percentages to the Standard input and output rates on the API pricing page and the model pages for GPT-6.1 Terra and GPT-6 Luna. Rates are per 1 million tokens for prompts up to 272K input tokens. The 2h and 8h columns are our calculations from OpenAI’s stated discounts.
| Model | Standard (input / output) | 2h window (30% off) | 8h window (40% off) | Batch API (50% off) |
|---|---|---|---|---|
| gpt-6.2-astra | $8.00 / $40.00 | $5.60 / $28.00 | $4.80 / $24.00 | $4.00 / $20.00 |
| gpt-6.2-sol | $2.00 / $10.00 | $1.40 / $7.00 | $1.20 / $6.00 | $1.00 / $5.00 |
| gpt-6.1-terra | $1.00 / $5.00 | $0.70 / $3.50 | $0.60 / $3.00 | $0.50 / $2.50 |
| gpt-6-luna | $0.10 / $0.50 | $0.07 / $0.35 | $0.06 / $0.30 | $0.05 / $0.25 |
The pricing page says the discount applies to “model token rates for eligible background requests.” The background guide adds that it applies to each eligible request, including follow-ups. The model pages we checked list higher rates for prompts above 272K input tokens, so check them before very large jobs. For the full rate card, see our OpenAI API pricing explainer.
How to request a discounted window
- Use the Responses API endpoint,
POST /v1/responses, with one of the four supported models. - Set
backgroundtotrue. - Set
background_options.completion_windowto2hor8h. - Set
service_tiertodefault. Discounted windows require Standard processing. Withauto, the request qualifies only if the API selects Standard. - Leave
streamunset or set it tofalse. - Save the response ID. Then either subscribe a webhook endpoint for your project to
response.completed(plus failed, incomplete or canceled events if you need them), or poll the retrieve endpoint while the status isqueuedorin_progress.
A minimal request body for an overnight document review looks like this, based on the guide’s example:
{
"model": "gpt-6.1-terra",
"input": "Review these documents for missing approvals: ...",
"background": true,
"background_options": { "completion_window": "8h" },
"service_tier": "default"
}
To go back to the undiscounted behavior, omit completion_window or set it to soonest. You can cancel an in-flight background response with POST /v1/responses/{id}/cancel. OpenAI notes that canceling twice is idempotent.
Multi-step agent runs and continuation IDs
The main difference from batch-style discounts is that background windows support agent workflows. OpenAI’s worked example uses gpt-6.1-terra with a function tool that retrieves a document by ID:
- Send the first request with an 8h window and a unique
background_options.continuation_idthat your application chooses for the run. - When the
response.completedwebhook arrives, deduplicate deliveries using thewebhook-idheader and check that the event’s response ID matches the one you saved. - Run the tool call the model asked for.
- Send a follow-up with the same model, service tier, background setting, completion window and continuation ID. Set
previous_response_idto carry forward history, and send the tool result as afunction_call_outputitem.
Reusing the window, continuation ID, model and project makes follow-ups eligible for higher queue priority. The continuation ID does not add prior history to the input by itself. If the earlier response has expired, OpenAI says to omit previous_response_id and replay the original input and every saved output item in order, including encrypted reasoning items. For more on agent-style builds, see our OpenAI Agents API guide.
Background windows vs Batch and Flex
OpenAI now has three ways to pay less for work that can wait. The Batch API gives a 50% discount, and each batch “completes within 24 hours (and often more quickly).” Its completion window can only be set to 24h. OpenAI recommends Batch for independent requests you can submit together up front, and background mode for multi-step runs.
Flex processing is priced at Batch rates but is a different trade. It uses service_tier: "flex", gives slower responses, and can return a 429 Resource Unavailable error, which is not charged. Flex is in beta with limited model availability; the pricing page lists GPT-6.1 Terra and GPT-6 Luna, but not the GPT-6.2 models, under Flex. Background windows, by contrast, use Standard processing, and once a 2h or 8h request is accepted the API automatically retries eligible capacity errors.
If you mainly use GPT-6.1 Terra, our GPT-6.1 Terra guide covers the model itself.
Limits and caveats
- Shared rate limit. 2h and 8h requests share the Batch/Background rate limit with Batch API jobs, and input tokens count at submission. The soonest window uses regular limits.
- No cache sharing. Standard requests outside the discounted windows cannot reuse prompt cache entries created by 2h or 8h requests, even for identical prompts.
- Temporary storage. Background mode stores response data temporarily even with
storeset to false or Zero Data Retention. OpenAI says this content is encrypted at rest for all three windows. - Targets, not guarantees. The guide describes the windows as completion targets, and says start times for follow-ups depend on scheduling.
- Beta status. The discounted windows are a public beta and cover only four models.
Frequently asked questions
What is OpenAI background mode?
Background mode runs a Responses API request asynchronously. You set background to true, save the response ID, and then poll the response or wait for a webhook. It suits long-running requests and multi-step agent runs where nobody is waiting on a live connection.
How much do the 2h and 8h completion windows save?
OpenAI says model token rates are 30% below Standard for the two-hour window and 40% below Standard for the eight-hour window. The default soonest window has no discount.
Which models support the discounted completion windows?
The public beta covers GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna. Requests must use Standard processing, so set service_tier to default.
Can I stream a 2h or 8h background response?
No. Streaming is only available with the default soonest window. For 2h and 8h requests, leave stream unset or set it to false, and use a webhook or polling to get the result.
Is the 8h window cheaper than the Batch API?
No. The Batch API costs 50% less than Standard and completes within 24 hours, while the 8h window is 40% below Standard. The windows suit agent runs with follow-up requests, while Batch suits independent requests you can submit together up front.
Sources
- OpenAI API docs: Background mode
- OpenAI API docs: Changelog (November 17, 2026 entries)
- OpenAI API docs: Pricing
- OpenAI API docs: Batch API
- OpenAI API docs: GPT-6.1 Terra model page
Image: OpenAI logo (symbol), public domain, via Wikimedia Commons.