OpenAI has changed how developers get cheaper API prices. For long-running jobs on its newest models, the Responses API now offers discounted completion windows in background mode: let a request finish within two hours and pay 30% less per model token, or allow eight hours and pay 40% less. The feature entered public beta on 17 November 2026 and replaces Flex processing on GPT-6.2 Astra and GPT-6.2 Sol.
Quick Take
- Two windows: “2h” costs 30% below Standard rates, “8h” costs 40% below Standard.
- Models: GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna, in public beta.
- No more Flex on new models: starting with GPT-6.2 Astra and Sol, new models launch without service_tier “flex.”
- Setup: background set to true, background_options.completion_window set to “2h” or “8h,” and service_tier set to “default.”
- Follow-ups: reusing the window, a continuation ID, model and project keeps the discount and gives follow-up requests higher queue priority.
In plain English
If your app does not need an answer right away, OpenAI will now give you a discount for waiting. Think of an overnight review of documents or a summary of the morning’s support tickets before an afternoon meeting. You tell the API you can wait up to two or eight hours, and you pay less for each token.
What happened
OpenAI’s API changelog entry for 17 November reads: “Choose a two-hour or eight-hour background completion window in the Responses API public beta. Model token rates are 30% lower than Standard for the two-hour window and 40% lower for the eight-hour window.”
The background mode guide adds that “starting with GPT-6.2 Astra and Sol, new models launch without Flex processing.” Requests to those models with service_tier “flex” return an error that points developers to background completion windows instead. The change landed alongside the launch of GPT-6.2 Astra and Sol.
How it works
- Target, not guarantee: the API targets completion of the initial response within the chosen window after it accepts the request.
- Standard processing only: service_tier must be “default.” With “auto,” a request qualifies only if the API happens to select Standard.
- Shared limits: 2h and 8h requests share the Batch and Background rate limit with Batch API jobs, and input tokens count at submission.
- No streaming: stream must be unset or false.
- Cache caveat: Standard requests outside the discounted windows cannot reuse prompt cache entries created by 2h or 8h background requests.
- Price example: GPT-6.2 Sol lists Standard input at $2.00 and output at $10.00 per million tokens, so the 8h window works out to $1.20 and $6.00.
Why it matters
For developers, background windows are a simple way to cut costs on batch-like work such as document review, evaluations or nightly reports, without rewriting jobs for the Batch API.
For teams that used Flex, the economics change. On older models Flex offered a bigger discount with response times closer to normal requests, and some developers in OpenAI’s community forum argue the new windows are a worse deal for interactive use. Budgets built on Flex pricing need a fresh look before moving to GPT-6.2.
For companies, it rewards planning. Work that can wait a few hours becomes up to 40% cheaper on model tokens.
What’s still open
- How reliably requests finish within the target windows at scale.
- Whether Flex returns for future models or is gone for good.
- When the beta becomes generally available and whether more models are added.
Sources
- OpenAI API changelog, 17 November 2026
- OpenAI API documentation, “Background mode” guide and pricing page
- OpenAI developer community forum, announcement thread on completion windows
i10x Perspective
OpenAI is nudging developers to separate urgent from patient work. For anyone running large evaluations or overnight pipelines, the 8h window is an easy saving. For anyone who relied on Flex for near-real-time apps on a budget, it is effectively a price increase. More from OpenAI and our Coding section.






