智能体
www.aibase.com ·
OpenAI 上线 GPT-Live-1 API：支持打断处理、工具调用与电话智能体
AIbase 报道称，OpenAI 正式向 API 推出新模型 GPT-Live-1，面向全双工实时语音应用。该模型将语音理解与语音输出整合在同一模型内，减少传统“语音转文字—大模型推理—文字转语音”链路带来的延迟，支持边听边说，并处理打断、停顿和背景噪声。开发者可通过系统提示调整语气、语速和对话风格，配置后端模型、工具与智能体框架，并支持原生语音识别与转写、字母数字理解、关键词偏置和轮次检测，可连接 Codex 等工具，通过 OpenAI Presence 开发查询企业系统、执行授权操作及转接人工的语音流程，应用于餐厅预订、客服等电话场景。报道称，语言学习平台 Speak 在接入后学习者思考停顿时的误打断减少近 80%；医疗服务平台借此简化架构并减少代码量。OpenAI 评测称，GPT-Live-1 在 Full Duplex Bench 上比 GPT-Realtime-2.1 高 30 个百分点，与 GPT-6Astra 在中等推理强度下组合，在 Tau3 端到端语音智能体任务测试中排名第一。上述型号与数据来自转述，尚待官方信息核实。
AI推荐理由 内容涉及 OpenAI 新模型 GPT-Live-1 API 的发布，包含全双工语音、打断处理、工具调用和电话智能体等具体能力，并给出 Speak 误打断减少约 80%、Full Duplex Bench 领先 GPT-Realtime-2.1 30 个百分点等评测数据，对开发者和语音 Agent 落地有参考价值。但来源为 AIbase 转述，非 OpenAI 官方原文，且 GPT-Live-1、GPT-Realtime-2.1、GPT-6Astra 等型号无法在输入中独立核实，评测数据也仅来自转述，需保留不确定性。
72
正文
OpenAI Officially Launches GPT-Live-1 API: Supports Interruption Handling, Tool Calls, and Phone Intelligent Agent
AIbase基地
Published inAI News · 4 min read · Sep 11, 2026
6
OpenAI has recently officially announced the introduction of the new model GPT-Live-1 to its API, providing core support for developers to create full-duplex real-time voice applications and business processes. The launch of this model marks the evolution of AI voice interaction from traditional "segmented processing" to true natural conversation.
In terms of core technical architecture, GPT-Live-1 achieves deep integration of speech understanding and speech output within the same model. It breaks through the multiple connection barriers in traditional models that involve "speech-to-text, large language model reasoning, and text-to-speech," thereby significantly reducing system latency. This model fully supports full-duplex conversation, allowing the system to listen to the user's voice while outputting speech, and accurately handle complex scenarios such as interruptions, pauses, and background noise during the conversation.
In addition to a smooth interactive experience, GPT-Live-1 also demonstrates high flexibility in function expansion and scenario implementation. Developers can easily adjust the model's tone, speed, and conversation style through system prompts, and configure backend models, tools, and intelligent agent frameworks. At the same time, the model supports native speech recognition and transcription, with capabilities including alphanumeric understanding, keyword bias, and round detection. It can connect with tools like Codex, and through OpenAI Presence, it can develop voice work that queries enterprise systems, performs approved operations, and transfers to human agents when necessary. In practical application scenarios, it can seamlessly integrate into phone conversations, handling full-duplex voice intelligent agent tasks such as restaurant reservations and customer service.
The effectiveness in practical applications is also significant. During early evaluations, the language learning platform Speak saw an almost 80% reduction in mis-interruptions during learners' thinking pauses after integrating GPT-Live-1, compared to previous round-based systems. Medical service platforms have also streamlined their architectures and significantly reduced code volume using this model. According to OpenAI's evaluation, GPT-Live-1 outperformed GPT-Realtime-2.1 by 30 percentage points in the Full Duplex Bench test, and ranked first in the Tau3 end-to-end voice intelligent agent task test when combined with GPT-6Astra at a moderate reasoning intensity.
GPT-Live-1AIvoiceinteractionOpenAIfull-duplexconversation
This article is from AIbase Daily
Scan to view
Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications.
—— Created by the AIbase Daily Team
© Copyright AIbase Base 2024, Click to View Source -