GPT-6.2 costs 30–40 per cent less in OpenAI’s API if the response can wait 2 or 8 hours
Developers can now choose a 2- or 8-hour target time for the first response in Responses API background mode, OpenAI said. A token then costs 30 or 40 per cent less than at the Standard tier. The feature is in public beta.
The OpenAI Developers account announced the discount on X on 18 November. The previous day, OpenAI had released its frontier model GPT-6.2 Astra and GPT-6.2 Sol, which is intended for everyday use. The discount applies to both models in OpenAI’s API.
The price depends on how quickly the response is needed. In background mode, an application does not need to keep a connection open while it waits for a response; it fetches the result later. In Responses API background mode, the developer sets completion_window to "2h" or "8h". With the two-hour window, a token costs 30 per cent less than at the Standard tier; with the eight-hour window, it costs 40 per cent less.
The window is a target time for the first response.
In OpenAI’s example, a user sends their meeting notes before the end of the working day. The user asks for a summary by the morning, using the eight-hour window. The image in the post shows the same request made with GPT-6.2 Sol.
OpenAI said the discount suits both single requests and multi-turn workflows that can wait. A developer can give the first request a continuation ID and use it in follow-up requests. The follow-up requests then get higher priority in the queue.
The table shows the Standard prices from which the discount is calculated:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $0.40 | $40 |
| GPT-6.2 Sol | $2 | $0.10 | $10 |
Prices per million tokens, from OpenAI’s launch announcement of 17 November 2026.
The posts do not specify whether the discount applies equally to all three price categories.
The cheaper time windows are in public beta. Developers enable them in Responses API background mode. OpenAI’s post links to more detailed instructions.
Read more: Models