Anthus AI

OpenAI has previewed a new Decisions API that returns a confidence score alongside model outputs, aimed at letting developers act on automated judgments rather than just raw text. Early commentary, including an Anthus AI analysis, argues the feature is only useful if those confidence numbers are genuinely well calibrated — and that untrustworthy scores could lead teams to make wrong decisions at scale. The debate centres on how reliably the model can quantify its own uncertainty.