OpenAI 正式发布 Decisions API,面向低延迟低成本小决策

来源:OpenAI:官网动态(RSS · 排除企业/客户案例) 2026年10月22日 03:00 AIHOT 评分:67 精选
摘要:OpenAI 宣布基于 GPT-6 Luna 的 Decisions API 正式可用,用于分类、路由、评分等有界答案问题,速度最高比通过 Responses API 使用 GPT-6 Luna 快 10x。
推荐理由:官方宣布 Decisions API 正式可用,给出与 GPT-6 Luna 的速度对比、定价和 GPT-Live 集成,便于评估路由分类类场景的收益。
# OpenAI 正式发布 Decisions API,面向低延迟低成本小决策 > 原标题:Build more responsive apps with the Decisions API - 来源:OpenAI:官网动态 - 发布时间:2026-10-22T19:00:00.000Z - AIHOT:https://aihot.news/items/xffhf8h0e0r145getmi9k1zgs - 原文:https://openai.com/index/decisions-api ## 摘要 OpenAI 宣布基于 GPT-6 Luna 的 Decisions API 正式可用,用于分类、路由、评分等有界答案问题,速度最高比通过 Responses API 使用 GPT-6 Luna 快 10x。 ## 推荐理由 官方宣布 Decisions API 正式可用,给出与 GPT-6 Luna 的速度对比、定价和 GPT-Live 集成,便于评估路由分类类场景的收益。 ## 正文 · 中文译文 ## 使用 Decisions API 构建响应更快的应用 为任何应用提供快速、专注、低成本决策。 分享 隆重推出 Decisions API [视频 1](https://vimeo.com/1238218282) ![Image 1: Decisions API — Video poster](https://images.ctfassets.net/kftzwdyauwt9/5eDGxJXBniEQU0wSAvC14M/9fefbe23f9c25e40458bdf6fe90da2c5/decisions-api-video-poster.jpg?w=3840&q=50&fm=webp) 00:00 05:25 应用和智能体在整个工作流中会做出许多小决策:如何对输入进行分类、将请求路由到哪里,或接下来采取哪个操作。单独来看,它们都很简单。但在规模化场景下,它们的延迟和成本会不断累积。 今天,我们正式全面推出 Decisions API。Decisions API 基于 [GPT‑6 Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/) 构建,将 Luna 的智能聚焦于答案有界的问题,帮助应用和智能体基于文本、图像和结构化输入做出近乎实时的决策,通过 Responses API 比 GPT‑6 Luna 快最多 10 倍。 自公开测试版以来,我们新增了原生 GPT‑Live 集成、针对重复使用上下文的提示缓存,以及美国和欧盟的推理驻留。开发者还可以传入原生 JSON 输入、在问题指令中包含图像,并且每个请求最多可提出 400 个问题,是此前上限的两倍。我们还持续改进延迟,包括在亚洲多个地区增加容量。 ## 团队如何使用 Decisions API Decisions 的早期采用情况表明,它适用于多种用例。 在 OpenAI 内部,各团队正在将 Decisions API 应用于产品和业务工作流。dots 团队用它来缩短响应时间并提升交互性。我们的个人财务团队用它来对交易描述进行分类,而我们的传播团队则用 Decisions 按主题对报道进行分类并评分,判断其是强化还是挑战关键叙事——帮助识别哪些内容值得进一步审查。 这些早期示例表明,快速、结构化的决策如何将传入信息转化为广泛用例中的有用行动。客户对 Decisions API 的应用同样广泛,从准确归因 AI 成本到让语言学习体验更具互动性。 第 1 张,共 5 张 > “\_借助 Box Extract 和\_\_Decisions API,我们对业务文档进行分类,并将其路由到正确的下一步。在我们的评估中,\__Decisions API 在 30 个保险证书案例上实现了约 5 倍更快的路由选择,准确率与\_\_Responses API 相同——自动化常规工作,并将置信度较低的案例发送给审核人员。_” _——Biswaroop Palit,Box 产品管理与集成平台副总裁_ > “_Decisions API 帮助我们在执行智能体生成的 \[ [rshell 命令⁠(在新窗口中打开)](https://github.com/datadog/rshell)\] 之前对其风险进行评分,从而自主修复问题。与可用的前沿模型相比,它以相当的准确率实现了快 40 倍的 p95——使 Decisions 能够置于修复操作的热路径中。_” _——AJ Stuyvenberg,Datadog 资深工程师_ > “_与 GPT‑6 Luna 相比,Decisions API 将我们的 AI 成本归因到实际工作的速度提升了 4 倍,成本降低了 30%,且不牺牲准确率。_” _——René Sultan,Ramp 研究工程师_ > “_在 Sierra,我们在实时客户对话中编排的部分环节使用 Decisions API 来选择正确的下一步操作。它以低于 280 毫秒的 p95 运行,比我们之前的方法快约 3 倍,每次调用成本降低 90%,且客户结果没有下降。_” _——Ajeet Grewal,Sierra 工程师_ > “对于语言学习者来说,时机至关重要。在我们的测试中,GPT‑Live 中的 Decisions API 集成能够将下一步行动的时间缩短 90% 以上,从 1.4 秒降至 0.13 秒,从而创造出更具互动性的学习体验。” —Andrew Hsu,Speak 首席技术官 - Box - Datadog - Ramp - Sierra - Speak - Box - Datadog - Ramp - Sierra - Speak ## 选择你的应用所需的决策类型 使用 Decisions API 非常简单:开发者定义一个问题和其允许的答案,然后以文本、图像或原生 JSON 形式提供上下文。该 API 会返回三种受支持的答案类型之一——概率、选择或评分——应用可据此决定接下来发生什么。多样的答案类型使 Decisions 能够适用于广泛的使用场景。 Decisions API 支持三种答案类型: | 类型 | 用途 | 主要结果 | | --- | --- | --- | | predicate | 估计某个条件为真的概率,例如产品是否有可见损坏。 | probability:条件为真的估计值,范围从 0 到 1。 | | choice | 从你预定义的选项中选择一个,例如智能体应当继续、重试还是请求澄清。 | choice:你所提供的值之一。响应中还包含概率和置信度。 | | score | 按照评分标准中的有序等级对输入进行评级,例如损坏的严重程度。 | score:各等级索引的概率加权平均值。 | ## 使用 Decisions API 和 GPT‑Live 构建实时语音体验 借助 [GPT‑Live](https://openai.com/index/introducing-gpt-live-1-in-the-api/),你可以为语音应用带来自然、富有表现力的对话,并具备流畅的打断处理能力,让对话保持连贯。Decisions API 现已[与 GPT‑Live 原生集成⁠(在新窗口中打开)](https://developers.openai.com/api/docs/guides/decisions-voice),帮助应用在实时对话中做出快速、聚焦的决策。客服智能体可以决定是否提出澄清问题、转接请求或升级问题。虚拟形象可以在用户对其说话时调整面部表情,并根据用户的话语以眨眼或微笑回应,从而创造更具互动性的体验。 下面的示例使用转录更新来选择虚拟形象的表情。将此配置添加到现有 GPT‑Live 会话中的 `session.triggers`。 #### JSON `1{2 "type": "decisions",3 "event_type": "session.input_transcript.delta",4 "decisions": {5 "questions": [{6 "type": "choice",7 "name": "reaction",8 "instructions": "Choose an appropriate reaction for the avatar based on the user's latest message.",9 "choices": [10 { "value": "neutral" },11 { "value": "happy" },12 { "value": "sad" },13 { "value": "concerned" }14 ]15 }]16 }17}` Live 会发出一个 `decision.event`;应用读取所选的 `reaction` 并相应地驱动角色动画。[了解更多关于将语音连接到 Decisions API 的信息。⁠(在新窗口中打开)](https://developers.openai.com/api/docs/guides/decisions-voice) ## 亲自试用 Decisions 与 GPT‑Live [音频 1](https://openai.com/index/decisions-api) 麦克风关闭 ![Image 2](https://openai.com/_next/static/immutable/media/neutral.0e2e43zp24lhc.webp) ![Image 3](https://openai.com/_next/static/immutable/media/thinking-bulb.1z1_50iiwd5ra.webp) 开始会话 在你开始之前麦克风处于关闭状态 正在连接 会话将在 0:00 后结束 会话已结束。如果你想继续交谈,请重新开始。 在你开始之前麦克风处于关闭状态 ## 表情面板 点击一张脸进行预览。 ![Image 4](https://openai.com/_next/static/immutable/media/curious.1z8d03c0o-3de.webp?w=96&q=90&fm=webp)好奇![Image 5](https://openai.com/_next/static/immutable/media/delighted.3o_81fgq98onp.webp?w=96&q=90&fm=webp)欣喜![Image 6](https://openai.com/_next/static/immutable/media/surprised.3rw2p6nwkx66y.webp?w=96&q=90&fm=webp)惊讶![Image 7](https://openai.com/_next/static/immutable/media/thinking-palette.1z5po-ddh-20f.webp?w=96&q=90&fm=webp)沉思![Image 8](https://openai.com/_next/static/immutable/media/empathetic.0mexjnh3muccz.webp?w=96&q=90&fm=webp)温柔![Image 9](https://openai.com/_next/static/immutable/media/mischievous.0piy4jm2de8n-.webp?w=96&q=90&fm=webp)俏皮![Image 10](https://openai.com/_next/static/immutable/media/neutral.0e2e43zp24lhc.webp?w=96&q=90&fm=webp)放松 ## 实时决策 观察 Decisions API 如何根据你们双方所说的话选择面部表情。 ### 你的话语 说话时观察反应。 ### 助手的话语 助手回复时观察面部表情。 试着说 我为我的猫排练了试镜。她中途走开了。 我有一个秘密。再靠近一点。 其实,是我妹妹得到了那个角色。这是她的首场演出! 我对明天感到紧张。你能陪我坐一会儿吗? 我刚拿到那个角色!你能和我一起庆祝吗? 做出惊讶的表情,然后告诉我你看到了什么。 你能帮我搜索下一次流星雨吗? 我为我的猫排练了试镜。她中途走开了。 此演示有时间限制。使用即表示你同意 OpenAI 的[条款⁠](https://openai.com/policies/terms-of-use/)并确认我们的[隐私政策⁠](https://openai.com/policies/privacy-policy/)。 ## 构建并评估你的第一个决策 Decisions API 现已通过专用的 Decisions 端点正式提供,并支持 GPT‑6 Luna。当前费率与缓存详情请参阅[定价与可用性指南⁠(在新窗口打开)](https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability)。 从你的应用经常做出的一个重复性决策开始。定义问题、可能的答案,以及区分这些答案所需的上下文。在[Decisions Playground⁠(在新窗口打开)](https://platform.openai.com/decisions)中用你自己的文本或图像,在反映真实使用情况的示例上进行测试,或按照[开发者指南⁠(在新窗口打开)](https://developers.openai.com/api/docs/guides/decisions)发出你的第一个请求。 Decisions API 的费用为每百万输入 token 0.10 美元。缓存输入的费用为每百万 token 0.01 美元。关于 30 分钟和 12 小时的缓存写入,[请阅读我们的定价指南。⁠(在新窗口打开)](https://developers.openai.com/api/docs/pricing?utm_source=chatgpt.com)你只需为输入 token 付费:没有输出 token 费用。区域处理可在美国和欧盟使用,符合条件的客户还可使用[零数据保留⁠(在新窗口打开)](https://developers.openai.com/api/docs/guides/your-data)。详情请参阅[定价与缓存⁠(在新窗口打开)](https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability)和[数据控制⁠(在新窗口打开)](https://developers.openai.com/api/docs/guides/your-data)。 - [API](https://openai.com/news/?tags=api) - [2026](https://openai.com/news/?tags=2026) ## 作者 OpenAI ## 正文 · 原文 ## Build more responsive apps with the Decisions API Fast, focused, low-cost decisions for any application. Share Introducing the Decisions API [Video 1](https://vimeo.com/1238218282) ![Image 1: Decisions API — Video poster](https://images.ctfassets.net/kftzwdyauwt9/5eDGxJXBniEQU0wSAvC14M/9fefbe23f9c25e40458bdf6fe90da2c5/decisions-api-video-poster.jpg?w=3840&q=50&fm=webp) 00:00 05:25 Applications and agents make many small decisions throughout a workflow: how to classify an input, where to route a request, or which action to take next. Individually, they are simple. At scale, their latency and cost add up. Today, we’re making the Decisions API generally available. Built with [GPT‑6 Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/), the Decisions API focuses Luna’s intelligence on questions with bounded answers, helping applications and agents make near real-time decisions from text, images, and structured inputs up to 10x faster than GPT‑6 Luna through the Responses API. Since public beta, we’ve added a native GPT‑Live integration, prompt caching for reused context, and inference residency in the US and EU. Developers can also pass native JSON inputs, include images in question instructions, and ask up to 400 questions per request, twice the previous limit. We’ve also continued to improve latency, including additional capacity in several regions in Asia. ## How teams are using the Decisions API Early adoption of Decisions has shown its applicability across a variety of use cases. Across OpenAI, teams are applying the Decisions API across product and business workflows. The dots team is using it to reduce response times and improve interactivity. Our Personal Finance team is using it to categorize transaction descriptions, while our Communications team is using Decisions to classify coverage by the themes and scores whether it reinforces or challenges key narratives—helping identify what warrants closer review. These early examples show how fast, structured decision-making can turn incoming information into useful action across a wide range of use cases. Customers are applying the Decisions API just as broadly, from accurately attributing AI costs to making language-learning experiences more interactive. 1 of 5 > “\_With Box Extract and\_\_the\_\_Decisions API, we classify business documents and route them to the right next step. In our evaluation,\__the\_\_Decisions API delivered ~5x faster route selection on 30 certificate of insurance cases, with the same accuracy as\_\_the\_\_Responses API—automating routine work and sending lower-confidence cases to reviewers._” _—Biswaroop Palit, VP, Product Management Platform and Integrations, Box_ > “_The\_\_Decisions API is helping us score the risk of agent-generated \[ [rshell commands⁠(opens in a new window)](https://github.com/datadog/rshell)\] before executing them to autonomously remediate issues. Compared with available frontier models, it delivered 40x faster p95 with comparable accuracy—making it possible to place Decisions\_\_in the hot path of a remediation action._” _—AJ Stuyvenberg, Staff Engineer, Datadog_ > “_The Decisions API has made it 4x faster and 30% cheaper to attribute our AI costs to the actual work when compared to GPT‑6 Luna, without sacrificing accuracy._” _—René Sultan, Research Engineer, Ramp_ > “_At Sierra, we use\_\_the\_\_Decisions API to choose the right next action during parts of our orchestration in live customer conversations. It runs at under 280 ms p95, about 3× faster and 90% lower cost per call than our previous approach, with no drop in customer outcomes._” _—Ajeet Grewal, Engineer, Sierra_ > “For language learners, timing matters. In our testing, the Decisions API integration in GPT‑Live was able to reduce the time to next action by over 90%, from 1.4s to 0.13s, creating a much more interactive learning experience.” —Andrew Hsu, CTO, Speak - Box - Datadog - Ramp - Sierra - Speak - Box - Datadog - Ramp - Sierra - Speak ## Select the type of decision your application needs Using the Decisions API is simple: developers define a question and its allowed answers, then provide context as text, images, or native JSON. The API returns one of three supported answer types—a probability, choice, or score—that applications can use to determine what happens next. The variety of answer types allows Decisions to be applicable across a wide variety of use cases. Decisions API supports three answer types: | Type | Use it to | Main result | | --- | --- | --- | | predicate | Estimate the probability that a condition is true, such as whether a product has visible damage. | probability: an estimate from 0 to 1 that the condition is true. | | choice | Select one of your predefined options, such as whether an agent should continue, retry, or ask for clarification. | choice: one of your supplied values. The response also includes probabilities and confidence. | | score | Rate an input against ordered levels in a rubric, such as the severity of damage. | score: the probability-weighted average of the level indices. | ## Build real-time voice experiences with the Decisions API and GPT‑Live With [GPT‑Live](https://openai.com/index/introducing-gpt-live-1-in-the-api/), you can bring natural, expressive conversation to your voice applications, with smooth interruption handling that keeps conversations flowing. Decisions API now [integrates natively with GPT‑Live⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/decisions-voice), helping applications make fast, focused decisions during live conversations. A customer support agent can decide whether to ask a clarifying question, route a request, or escalate an issue. An avatar can adjust its facial expressions as a user speaks to it, and based on the user’s words, respond with a wink or a smile to create a more interactive experience. The example below uses transcript updates to select an avatar’s expression. Add this configuration to `session.triggers` in an existing GPT‑Live session. #### JSON `1{2 "type": "decisions",3 "event_type": "session.input_transcript.delta",4 "decisions": {5 "questions": [{6 "type": "choice",7 "name": "reaction",8 "instructions": "Choose an appropriate reaction for the avatar based on the user's latest message.",9 "choices": [10 { "value": "neutral" },11 { "value": "happy" },12 { "value": "sad" },13 { "value": "concerned" }14 ]15 }]16 }17}` Live emits a `decision.event`; the application reads the selected `reaction` and animates the character accordingly. [Learn more about connecting voice to Decisions API. ⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/decisions-voice) ## Try Decisions with GPT‑Live for yourself [Audio 1](https://openai.com/index/decisions-api) Mic off ![Image 2](https://openai.com/_next/static/immutable/media/neutral.0e2e43zp24lhc.webp) ![Image 3](https://openai.com/_next/static/immutable/media/thinking-bulb.1z1_50iiwd5ra.webp) Start session Microphone off until you start Connecting Session ends in 0:00 Session ended. Start again if you'd like to keep talking. Microphone off until you start ## Expression palette Tap a face to preview it. ![Image 4](https://openai.com/_next/static/immutable/media/curious.1z8d03c0o-3de.webp?w=96&q=90&fm=webp)Curious![Image 5](https://openai.com/_next/static/immutable/media/delighted.3o_81fgq98onp.webp?w=96&q=90&fm=webp)Delighted![Image 6](https://openai.com/_next/static/immutable/media/surprised.3rw2p6nwkx66y.webp?w=96&q=90&fm=webp)Surprised![Image 7](https://openai.com/_next/static/immutable/media/thinking-palette.1z5po-ddh-20f.webp?w=96&q=90&fm=webp)Pensive![Image 8](https://openai.com/_next/static/immutable/media/empathetic.0mexjnh3muccz.webp?w=96&q=90&fm=webp)Tender![Image 9](https://openai.com/_next/static/immutable/media/mischievous.0piy4jm2de8n-.webp?w=96&q=90&fm=webp)Cheeky![Image 10](https://openai.com/_next/static/immutable/media/neutral.0e2e43zp24lhc.webp?w=96&q=90&fm=webp)At ease ## Live decisions Watch as the Decisions API chooses a facial expression based on what each of you says. ### Your words Watch the reaction as you speak. ### Assistant’s words Watch the face as the assistant replies. Try saying I rehearsed my audition for my cat. She walked out halfway through. I have a secret. Lean in a little closer. Actually, my sister got the part. It’s her first show! I’m nervous about tomorrow. Can you sit with me? I just got the part! Can you celebrate with me? Look surprised, then tell me what you saw. Could you search the web for the next meteor shower? I rehearsed my audition for my cat. She walked out halfway through. This demo is time-limited. By using it, you agree to OpenAI's [Terms⁠](https://openai.com/policies/terms-of-use/) and acknowledge our [Privacy Policy⁠](https://openai.com/policies/privacy-policy/). ## Build and evaluate your first decision Decisions API is generally available today with GPT‑6 Luna through the dedicated Decisions endpoint. See the [pricing and availability guide⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability) for current rates and caching details. Start with one recurring decision your application makes often. Define the question, the possible answers, and the context needed to distinguish them. Test it on examples that reflect real usage, in the [Decisions Playground⁠(opens in a new window)](https://platform.openai.com/decisions) with your own text or images, or follow the [developer guide⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/decisions) to make your first request. Decisions API costs $0.10 per million input tokens. Cached input costs $0.01 per million tokens. For 30-minute and 12-hour cached writes, [read our pricing guide.⁠(opens in a new window)](https://developers.openai.com/api/docs/pricing?utm_source=chatgpt.com) You only pay for input tokens: there are no output-token charges. Regional processing is available in the US and EU, and [Zero Data Retention⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/your-data) is available to eligible customers. See [pricing and caching⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability) and [data controls ⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/your-data)for details. - [API](https://openai.com/news/?tags=api) - [2026](https://openai.com/news/?tags=2026) ## Author OpenAI