What actually happened
OpenAI released the Decisions API in public beta on 6 October 2026. The dedicated POST /v1/decisions endpoint evaluates text, images or both and returns typed answers: a predicate probability, a choice from a fixed list with confidence, or a score against ordered levels. The currently supported model is GPT-6 Luna. OpenAI says a comparable decision can be returned about ten times faster than through the Responses API.
The guide describes input-only billing, with no output-token charge for decision answers and a base rate of $0.10 per million input tokens, subject to applicable multipliers. Image inputs can be inline base64 data URLs or publicly accessible HTTP(S) URLs; file IDs are not supported. This is a correction to earlier guidance that said images had to be base64 only.
Why a small team cares
A solo consultant scoring incoming leads or an agency routing support tickets may currently ask a model for a category and then parse the prose answer. A typed result removes that extraction step and exposes a probability the workflow can use directly. The same pattern supports prioritizing messages, classifying content, or checking image-based inputs.
Test 50 to 100 cases already labeled by your team. Compare the Decisions answer with the old workflow, including edge cases and false positives. Set the routing threshold from actual observed quality rather than trusting the returned confidence as an automatic approval for high-impact actions.
Hype vs useful
The published speed and pricing claims make Decisions attractive for narrow classification work, but this beta is not a general replacement for Responses. It does not generate drafts or explanations, and model decisions still need empirical accuracy checks. Input size, regional charges and evolving beta behavior can affect costs. One measured migration is justified; a full rebuild is not.
Source: OpenAI API (https://developers.openai.com/api/docs/guides/decisions)