IT之家 9 月 30 日消息，在今天召开的 2026 开发者日活动中，OpenAI 宣布推出 Decisions API，这是一个面向实时、低延迟分类与路由情形的全新 API，可以大约在 150 毫秒内返回结果，比借助常规 API 使用 Luna 快大约 10 倍。
定位维度，Decisions API 为“快速单次决策”外观，把模型（当前基于 OpenAI 的小型模型 Luna）的“智能”聚焦到一组使用者预定义的难题 + 有限答案集合上，用极低的延迟返回结构化决策结果。
在当下同类模型中，主要竞争模型来自 TypeSafe 的 Jev，不过 Decisions API 专注于将模型智能转化为快速决策。
开发者给出上下文（文本或图片），并预先定义好“难题 + 可选答案”，API 返回选中的答案及置信度等结构化信息，用于后续业务逻辑。
在典型应用情形维度，Decisions API 比较合适以下情形：
本文来源：https://www.ithome.com/1/008/531.htm
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
IT之家 News, September 30: At today's 2026 Developer Day event, OpenAI announced the launch of the Decisions API, a new API for real-time, low-latency classification and routing scenarios. It can return results in approximately 150 milliseconds, about 10 times faster than using Luna via the regular API.
In terms of positioning, the Decisions API is designed as a "fast single-shot decision" interface, focusing the model's (currently based on OpenAI's small model Luna) "intelligence" on a set of user-predefined problems plus a limited answer set, returning structured decision results with extremely low latency.
Among current comparable models, the main competitor is TypeSafe's Jev, but the Decisions API focuses on turning model intelligence into fast decisions.
Developers provide context (text or image) and predefine "problem + possible answers"; the API returns structured information such as the selected answer and confidence score for downstream business logic.
In terms of typical application scenarios, the Decisions API is well suited for the following cases: