OpenAI’s API changelog marks the Decisions API generally available on October 22 with gpt-6-luna. The API is designed to turn text, images and structured inputs into typed decisions; the per-request limit has risen from 200 to 400 questions. It also accepts structured input_json and can share cached input across questions. For a system like Wake Up, that invites a narrowly defined experiment: score candidate stories for actionability, personal relevance, duplication and tone before the editorial selection is made.
The potential benefit is not simply a faster model. A single well-defined classification pass could make editorial rules visible and auditable, producing consistent High/Medium/Low and Hopeful/Neutral/Concerning labels. But there is no evidence yet that the Decisions API would improve research quality, source verification or writing. It should be compared with your current method on a held-out set of stories rather than dropped into production.
Start with twenty already published candidates and their known editorial classifications. Measure accuracy, latency and billed tokens, and inspect the mistakes. Keep the MiniMac publishing process unchanged. A classification tool that costs pennies and catches repeat stories could be useful; one that produces confident but unreliable labels is not.
What to do: Prototype a 20-story Decisions API classification test against known Wake Up labels and compare accuracy and cost before changing production.