OpenAI Responses API adds 2-hour and 8-hour completion windows in background mode
BIBIBI
AT A GLANCE
PANews reports that OpenAI has added 2-hour and 8-hour completion windows in background mode, with token rates reduced by 30% and 40% respectively, but subject to usage restrictions.
Article
OpenAI launches background mode: Long tasks can run for 8 hours, with token costs reduced by up to 40%
PANews reported on November 20 that, according to OpenAI developer documentation, OpenAI has added a "completion window" option to the Responses API's background mode. When selecting the 2h or 8h window, model token rates are 30% and 40% lower than Standard, respectively; the default soonest window offers no discount. This feature is currently in public beta on GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, and GPT-6 Luna. The documentation also reveals that starting with GPT-6.2 Astra and Sol, new models no longer offer Flex processing. Discounted windows require Standard processing, do not support streaming output, and share rate limits with the Batch API. Subsequent requests using the same continuation_id can achieve higher queue priority...
PANews reported on November 20 that, according to OpenAI developer documentation, the Responses API has added a "completion window" option in background mode.
02
The report notes that when choosing the 2h or 8h window, the model token rates are 30% and 40% lower than Standard, respectively; the default soonest window is not discounted.
03
The report lists the public test models as GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, and GPT-6 Luna.
04
The report notes that, starting with GPT-6.2 Astra and Sol, the new models no longer offer Flex processing.
05
The report notes that the discount window requires Standard processing, does not support streaming output, and shares rate limits with the Batch API.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
According to the report, users can choose a longer completion window for background tasks in exchange for a lower model token rate. A 2-hour window corresponds to a 30% discount, and an 8-hour window corresponds to a 40% discount; the default window has no discount. The discount scheme cannot use streaming output and has processing method and rate limit requirements.
Why it matters to readers
This information concerns the pricing and usage conditions of background tasks: when comparing discounts, you must also consider the completion window, output method, and rate limits; do not focus solely on the discount amount.
If you use these APIs, you should pay particular attention to whether your task is suitable for a longer completion window and whether streaming output is required; the discounts mentioned in the report are not directly applicable to all requests.
Risks and unknowns
The material is a paraphrase of the developer documentation by PANews; the original official documentation was not provided, and the related features, models, and pricing have not yet been independently verified.
The main text is truncated at the continuation_id and queue-related descriptions, making it impossible to confirm the complete rules for subsequent requests.
Is the statement “long tasks can stay active for 8 hours” in the title exactly equivalent to “8h completion window” in the text? No explanation is provided in the material.
Related Developments
Loading event timeline…
Related concepts
Completion window
This term is not in the glossary yet. Browse related concepts in the glossary.