Skip to content
Neural Daily

ToolsNews

Responses API: 40% off if you can wait 8 hours

🇭🇺 Olvasd el magyarul →

OpenAI's Responses API now offers 2-hour and 8-hour completion windows in public beta, with model token rates 30% and 40% below Standard. New models launch without Flex processing, so this is the new way to save on non-urgent work.

· 2 min read

Az OpenAI Developers dokumentáció borítóképe: API Docs, Background mode felirat rózsaszín és narancs háttéren
Image: OpenAI · Screenshot · Source

What's new

Since November 17, the Responses API background mode lets you pick a completion window (OpenAI). The background_options.completion_window field accepts soonest (the default, no discount), 2h or 8h. Model token rates are 30% below Standard for 2h and 40% below for 8h (OpenAI). The API targets completion within the window, counted from when it accepts the request.

  • Supported models: GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna.
  • Example: GPT-6.2 Astra Standard is $8 input and $40 output per million tokens; with the 8h window that becomes $4.80 and $24.
  • Requirements: background: true, Standard processing (service_tier: default), no streaming.
  • 2h and 8h requests share the Batch/Background rate limit with Batch API jobs.

It works for agents too

The discount also applies to follow-up requests, which matters for tool-calling flows (OpenAI). If the first response asks your app to fetch a document, you send the tool result in a new request. Reusing the same window, continuation_id, model and project gives the chain higher queue priority. The continuation_id does not add prior history to the input; it only links related requests.

Watch out for

  • Starting with GPT-6.2 Astra and Sol, new models launch without Flex processing (service_tier: flex). If you relied on Flex to save money, this is the replacement.
  • With service_tier: auto, a request qualifies only if the API happens to select Standard.
  • Standard requests outside the discounted windows cannot reuse prompt cache entries created by 2h or 8h requests, even for identical prompts.

Who it's for

Anything nobody is waiting on live: overnight document processing, morning ticket summaries, eval runs, data labeling. The Batch API was the obvious choice until now, but it is file-based with separate job management. This is the same Responses API call with one extra field, and it covers tool-calling agent chains as well. Combined with the earlier 12-hour prompt cache (OpenAI API adds 12-hour prompt caching), non-urgent workloads can get noticeably cheaper.

    By Mikola Máté Patrik

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles