OpenAI adds two- and eight-hour background completion windows to Responses API beta
OpenAI's Responses API public beta lets developers choose a two-hour or eight-hour background completion window, with lower token rates for each.
OpenAI's API changelog for November 17, 2026 says the Responses API public beta now lets developers choose a two-hour or eight-hour background completion window. Model token rates are 30% lower than Standard for the two-hour window and 40% lower for the eight-hour window. The entry lists gpt-6.2-astra, gpt-6.2-sol, gpt-6.1-terra and gpt-6-luna as the models it covers. The changelog gives no further detail on how the windows behave.
Background mode lets a developer submit a Responses API request and continue without waiting on the connection. The application sets background to true, saves the response ID the API returns, and retrieves the result later. The completion window is set through background_options.completion_window. The default is the soonest option, which carries no model token discount. The two discounted windows are the 2h and 8h options. The background mode page pairs each window with a use case: the two-hour window for analyzing a morning's support tickets before an afternoon triage meeting, and the eight-hour window for reviewing a batch of documents overnight.
The window is a target, not a guarantee. The page says that for 2h and 8h, the API targets completion within the selected window after it accepts the request. The clock starts at acceptance, not when the developer sends the call. The discount applies to model token rates on each eligible request, including follow-ups. That matters for agent workflows, where an application runs a tool and sends its output in a second request. The page's example is an eight-hour request that asks the application to look up a document, followed by a follow-up request with the result. It says both requests qualify for the 40% discount. Reusing the same window, continuation ID, model and project qualifies follow-ups for higher queue priority. After the API accepts a two-hour or eight-hour request, it automatically retries eligible capacity errors.
Background mode also stores response data temporarily, even when a developer sets store to false or uses Zero Data Retention. The API uses that data to run the request and make the result retrievable. For two-hour and eight-hour requests, the page says the API encrypts the temporarily stored request content at rest. The page doesn't say what happens to a request that misses its target window, or whether the discount changes in that case. That is the detail a team would need before pricing a workload around the eight-hour option.
For receiving results, the page recommends a webhook subscribed to response.completed, along with events for failed, incomplete or canceled responses the developer wants to handle. The application should verify the signature on each incoming event, hand longer work to a background worker, and return a 2xx response promptly. Developers who can't receive webhooks can poll the response by its ID instead.
What changed
The Responses API public beta now lets developers choose a two-hour or eight-hour background completion window, with model token rates 30% lower than Standard for the two-hour window and 40% lower for the eight-hour window.
Type: Pricing and limits · Takes effect: Tuesday, November 17, 2026 · Applies to: Responses API, gpt-6.2-astra, gpt-6.2-sol, gpt-6.1-terra, gpt-6-luna
Why it matters
Teams that can wait for results could cut token costs by accepting a longer completion window rather than paying for immediate responses. The discount applies to the windows OpenAI lists, so the saving depends on how long a job can tolerate waiting.
What you can do
Identify non-urgent workloads that could tolerate a two- or eight-hour wait and compare their token costs under the lower background rates.
Original source
Where we saw it
OpenAI API changelog
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works