AI Card Bao WBCXAFGBMGDEVELOPER.AICARDBAO.COM

Industry News

OpenAI background mode public beta: If you are willing to wait 8 hours for the task, the token is 40% cheaper

OpenAI has offered a “more affordable” price for this API.

The method is to set the `background` parameter to `true` in the request, and then use `background_options.completion_window` to specify a completion time limit. Choosing a 2-hour deadline results in a 30% discount on the token fee compared to the standard rate; selecting an 8-hour deadline yields a 40% reduction.

Currently, the public beta version includes four models: GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, and GPT-6 Luna.

Let’s take Astra as an example. The standard price is $8 per million tokens; therefore, the cost for an 8-hour period would be $4.8 or $24, depending on the specific configuration.

There are three areas where one is prone to tripping.

Firstly, if the parameter `completion_window` is not specified, the default value “soonest” will be used, and the task will still be executed asynchronously – however, in this case, no discount will be applied at all. Just because the option “background” is selected does not mean that costs will be reduced; upon checking the bill, you will discover that the actual price remains the same as the original one.

Secondly, the background mode will temporarily store response data. Even if you set “store” to “false”, or even if your account is configured to retain no data at all, the data will still be stored. For teams with strict requirements regarding data compliance, this aspect must first be approved by the legal department.

Thirdly, the time limit is a goal, not a guarantee. The wording in the document states that “the goal is to complete the task within the selected time frame,” but it does not specify how charges will be calculated in case of a timeout.

It is situated between two existing options. The Batch API has always offered a 50% discount, but it only allows requests to be submitted in the form of files within a 24-hour time limit; it does not support the use of any additional tools during this process. The Background Mode is slightly cheaper, but it requires the use of the Responses API instead. With this mode, you can use various tools and run multiple iterations of your tasks, making it ideal for intelligent agent programs that need to operate throughout the night.

Typical usage example: Submit a batch of contracts for review before leaving work, select 8 hours to process them, and retrieve the results the next morning; for the task of analyzing morning work orders in order to attend an afternoon meeting, allocate 2 hours.

This function is not useful in scenarios where people stare at a screen waiting for results. The intelligent agents used for coding and customer service interactions still come at the same old prices.

Related